Working Paper Number 114
March 2007
Do visas kill? Health effects of African
health professional emigration
By Michael Clemens
Abstract
The emigration of highly skilled workers can in theory lower social welfare in the migrant-sending
country. If such workers produce a good whose consumption conveys a positive externality—such as
nurses and doctors in a very poor country—the loss can be greater, and welfare can even decline
globally. Policies to impede emigration thus have the potential to raise sending-country and global
welfare. This study uses a new database of health worker emigration from Africa to test whether
exogenous decreases in emigration raise the number of domestic health professionals, increase the
mass availability of basic primary care, or improve a range of public health outcomes. It identifies
the effect through two separate natural quasi-experiments arising from the colonial division of the
African continent. These produce exogenous changes in emigration comparable to those that would
result from different immigration policies in principal receiving countries. The results suggest that
Africa's generally low staffing levels and poor public health conditions are the result of factors
entirely unrelated to international movements of health professionals. A simple model proposes that
such results would be explained by segmentation of health workforce labor markets in the sending
countries. The results further suggest that emigration has caused a greater production of health
workers in Africa.
The Center for Global Development is an independent think tank that works to reduce global poverty and
inequality through rigorous research and active engagement with the policy community. Use and
dissemination of this Working Paper is encouraged, however reproduced copies may not be used for
commercial purposes. Further usage is permitted under the terms of the Creative Commons License. This
research was generously supported by the John D. and Catherine T. MacArthur Foundation and by the
William and Flora Hewlett Foundation. The views expressed in this paper are those of the author and
should not be attributed to the directors or funders of the Center for Global Development.
www.cgdev.org
________________________________________________________________________
Do visas kill?
Health e¤ects of African health professional emigration
Michael A. Clemens
Center for Global Development
March 13, 2007
Abstract
The emigration of highly skilled workers can in theory lower social welfare in the migrant-sending
country. If such workers produce a good whose consumption conveys a positive externality— such as
nurses and doctors in a very poor country— the loss can be greater, and welfare can even decline globally.
Policies to impede emigration thus have the potential to raise sending-country and global welfare. This
study uses a new database of health worker emigration from Africa to test whether exogenous decreases
in emigration raise the number of domestic health professionals, increase the mass availability of basic
primary care, or improve a range of public health outcomes. It identi…es the e¤ect through two separate
natural quasi-experiments arising from the colonial division of the African continent. These produce exogenous changes in emigration comparable to those that would result from di¤erent immigration policies
in principal receiving countries. The results suggest that Africa’s generally low sta¢ ng levels and poor
public health conditions are the result of factors entirely unrelated to international movements of health
professionals. A simple model proposes that such results would be explained by segmentation of health
workforce labor markets in the sending countries. The results further suggest that emigration has caused
a greater production of health workers in Africa.
JEL codes F22, I18, O15. Correspondence to mclemens@cgdev.org. Mailing address: Center for Global Development, 1776
Massachusetts Ave. NW, Suite 301, Washington, DC 20036-1909 USA. Tel. (+1) 202-416-0722, Fax (+1) 202-416-0750. This
research was generously supported by the John D. and Catherine T. MacArthur Foundation and by the William and Flora
Hewlett Foundation. The opinions expressed here are those of the author alone and do not necessarily re‡ect those of the Center
for Global Development, its board, or its funders. Gunilla Pettersson and Sami Bazzi provided superb research assistance. I
would like to thank Alok Bhargava, Nancy Birdsall, Frédéric Docquier, Tamara Fox, Ruth Levine, Prachi Mishra, Todd Moss,
Ça¼
g lar Özden, and seminar participants at MIT, the Northeast Universities Development Conference, the University of the
Witswatersrand, and the Sydney G20 summit. Any errors, however, belong to the author.
1
Introduction
Do international borders raise welfare? In what conditions would a benevolent social planner choose less
permeable frontiers over more open ones? These questions arguably spawned the …eld of economics: Hume,
Smith, and Ricardo built some of the foundational models of the discipline to ask how much their small
island nation should interfere in cross-border goods trade, and The Economist newspaper emerged in 1843
directly from an anti-tari¤ society. Debates on intervention in other international factor ‡ows— capital and
labor— followed, but lost priority after both crashed lastingly after 1914. Only in recent years did both
‡ows regain their belle époque levels worldwide, and the debates are back. Many new proposals to restrict
international labor movements arise from a facet of the new upsurge in migration that has no precedent: the
international mass migration of highly skilled workers.
When returns to skill were smaller and migration comprised primarily low-skill workers, welfare analysis
focused on the destination countries. Those countries remain the focus of the large bulk of migration
research. But the shift toward substantial movements of skilled workers has generated a large theoretical
and policy literature on the sending-country consequences of migration. Low-income countries need skilled
professionals to develop, a common argument goes, and nearly all developing-country governments subsidize
professional education in order to reap positive externalities from those engineers’ ideas, those lawyers’
contracts, those physicians’ care. But a thicket of labor and credit market failures prevents them from
being paid their marginal social product. If they departed for countries where their private gain better
re‡ects their contribution, they would be lost (“brain drain”) and a public good would become a private
good (“poaching”). Government intervention would be justi…ed, by this reasoning, to ensure that the positive
externality is captured by the country of origin. Warranted public acts would include contractual restrictions
on emigration (in self-interested sending countries) and bans on recruitment or tight profession-speci…c
immigration quotas (in altruistic receiving countries).
A related case has been made in the literature on other factor movements. Volumes of theoretical research
have argued that subsidies to infant industries raise welfare due to positive scale and network externalities
in production. Here, too, the externality raises the possibility that restricting international factor ‡ows can
raise welfare; a more impermeable border makes everyone better o¤. Theory leaves ambiguous the degree and
duration of the proper intervention, however, whence a massive empirical literature testing whether in fact
countries with borders more impermeable to goods trade are in fact better o¤. This empirical literature was
made possible by data; since 1948 the International Monetary Fund has published detailed annual, bilateral
data on international goods trade.
1
But no such data have existed for skilled-worker movements, so it has been impossible to test equally
plausible hypotheses about the sending-country welfare consequences of international skilled worker movements. There is no place to look up, say, how many physicians or engineers moved from Cameroon to Quebec
in 1997, much less how many went from Quebec to Cameroon. This paper makes the question tractable by
choosing one region and one profession where welfare consequences of emigration are particularly plausible:
African health professionals. It uses a novel database collected for this investigation to test whether or not
greater net emigration of such workers, all else equal, reduces sending-country welfare.
It begins with a simple welfare model from which emerge two natural quasi-experiments we can use to
test the sending-country welfare consequences of exogenous changes in the permeability of borders to health
professional emigration. Both quasi-experiments relate to the colonial division of the African continent.
The e¤ects are tested via cross-country regressions, whose use has been much maligned in recent years as
being far more convenient than theoretically appropriate to questions such as elucidating the determinants
of growth. In the present case, however, cross-country hypothesis testing is precisely the theoretically correct
framework. International migration policy is applied exclusively at international borders, so no other unit
of analysis besides countries would be meaningful.
The results presented here fail to detect any negative impact of even massive movements of health
professionals out of Africa upon health worker stocks, basic primary health care availability, and public
health outcomes in African migrant-sending countries. They suggest that the absence of a negative e¤ect of
emigration on domestic health worker stocks is plausibly the result of a positive elasticity of labor supply to
the expected wage increases a¤orded by the emigration option. They furthermore suggest that the absence
of an e¤ect of emigration on basic primary care availability and public health conditions may be due to
the determination of these outcomes primarily by factors unrelated to cross-border movements of health
professionals. These include the geographic and public/private sector distribution of health professionals in
the sending country, the skill mix of the health workforce, performance incentives faced by health professionals
in the sending country, and the relative importance of primary care versus prevention to the disease burden
in Africa.
2
Academic and policy debates in need of evidence
General equilibrium models in which the sole market failure is the international border predict large, positive
global welfare gains from opening the border (e.g. Hamilton and Whalley 1984; Moses and Letnes 2004;
2
Walmsley and Winters 2005). Within the sending countries, the simplest static models predict unambiguous
declines in welfare (though always exceeded by gains to migrants). More recent and nuanced models that
account for changing returns to other factors, remittances, and other forces predict increased welfare in some
sending countries from broad categories of movement comprising all “unskilled”or “skilled”labor collectively.
For decades, however, a strand of strictly theoretical literature has debated whether the migration of certain classes of skilled workers must always decrease sending-country welfare due to a positive local external
e¤ect of the exercise of their professions. Grubel and Scott (1966, 1968) argued that even the emigration
of physicians from developing countries could raise sending-country welfare due to technology transfer, remittances, human capital formation induced by the migration opportunity, and other forces. Aitken (1968)
…ercely contested that this “amazing conclusion ... is arrived at only because of an error in [Grubel and
Scott’s] analysis”, positing that large (especially dynamic) external e¤ects mean the welfare of the sendingcountry population must decrease with the departure of each and every physician.
Since then, numerous papers have re…ned these qualitative theoretical arguments with careful modeling
of the relevant externalities (e.g. Miyagiwa 1991). Several have focused on precisely modeling the e¤ect of
emigration opportunities on human capital investment (Mountford 1997; Stark et al. 1997, 1998; Beine et
al. 2001).
This discussion has been entirely theoretical, with three recent exceptions. Beine et al. (2001) show
that aggregate migration rates are correlated with overall schooling in a small cross-section of countries.
With greatly improved data, Beine et al. (2003) …nd a causal e¤ect of emigration among the tertiaryeducated on the sending-country stock of tertiary educated. They produce plausibly exogenous variance
in emigration rates using four instrumental variables: sending-country population density, total population,
“racial tensions”, and the stock of migrants at destination. Finally, Bhargava and Docquier (2006) assemble
novel data on physician emigration from 39 African countries and test its causal e¤ect on life expectancy
and adult deaths due to AIDS in the sending countries.
That last study di¤ers from the present one in several ways. First, it counts physicians abroad for each
African country by an odd blend of country of birth, country of training and country of citizenship. These
three measures often give radically di¤erent measures of physician out‡ow, as described in Clemens and
Pettersson (2006). Second, and very importantly, its identi…cation strategy allows no way to check for weak
instruments— a problem proven capable of generating spurious inference by several recent studies. Third,
although it represents a tireless attempt to collect panel data, the majority of the migration data in the
panel are not observed but imputed. The African physician stocks reported there occur at …ve points in time
3
for 16 receiving countries, but in 10 of those countries, only one or two timepoints are actually observed; the
rest are interpolated. Since migrant stocks are stable over time this amounts to roughly doubling the size
of the dataset without adding any new empirical information— arti…cially shrinking standard errors with a
substantial danger of spurious inference. Fourth, that study tests whether or not physician emigration has
e¤ects on AIDS deaths without testing the most obvious mechanism of that e¤ect: whether or not physician
emigration in fact decreases the number of physicians in the sending country.
Bhargava and Docquier …nd that the fraction of physicians abroad has a positive and signi…cant e¤ect on
the number of adult deaths due to AIDS in general, while— interestingly— it has a negative and signi…cant
e¤ect on AIDS deaths in countries where HIV prevalence is low. They also test the e¤ect of the fraction
of physicians abroad on life expectancy in the sending country, and …nd no e¤ect (a positive coe¢ cient
statistically indistinguishable from zero).
The present paper extends this literature in two ways. First, it makes a small theoretical contribution
by pointing out that the emigration of a health worker does not constitute the loss of a public good to the
extent that trained health professionals in the sending country work in a sector with limited external e¤ect
(the urban sector, the private health sector, and the non-clinical sector). This subtlety takes on tremendous
importance in the least-developed economies, where domestic labor market failures loom large. Second, and
more importantly, it assembles novel data to create a simple and transparent empirical test of the welfare
e¤ects of an exogenous change in the emigration of African physicians and professional nurses. It uses a
single, consistent de…nition of the emigration rate, a clear and transparent identi…cation strategy, and real
observed data.
These debates are not at all purely academic; government policies across the world have arrived well in
advance of clear evidence on their e¤ects. On the sending-country side, recent South African legislation has
sought to discourage nurse emigration with a combination of punitive licensing measures for emigrants and
obligatory in-country service for new graduates. On the receiving country side, in 2001 the United Kingdom’s
National Health Service progressively banned direct public recruitment of health professionals from all but
two developing countries and has pressured private agencies (and other countries) to follow suit. The welfare
consequences of these policies in the sending countries remain unknown.
3
The e¤ects of emigration are theoretically ambiguous
In Hamilton and Whalley’s (1984) classic model of international migration, the elimination of international
borders unambiguously results in 1) a smaller labor supply in low-wage areas, 2) lesser welfare in the sending
4
country, 3) greater welfare to migrants and receiving countries, and 4) increased global welfare. This section
extends that model to incorporate a positive externality from a good produced by the workers who move.
Here that good is “health”, and greater average consumption of health increases the incomes of everyone
in the country. The marginal social product of health professionals’labor thus exceeds the private product.
With restrictive assumptions, increased openness to migration still lowers sending-country welfare but can
now actually lower global welfare. Under alternative conditions, the change in sending-country labor supply
and welfare— as well as global welfare— can be small or even positive.
3.1
3.1.1
Demand for health
Competitive equilibrium with a positive income externality from health
A representative agent consumes two goods: x and h (“health”), which generate utility through the Constant
Elasticity of Subsitution (CES) function. Suppose that a portion of the agent’s income is a linear, separable
function of her consumption of health relative to others. If others consume more health than she does, her
income goes up; the more the agent’s neighbors spend on malaria prevention, for example, the less likely she
is to contract the disease and miss work. But if others consume less health than she, her income goes down.
In other words income m = m +
of h by other agents, and
h
h , where m is some base level of income, h is average consumption
> 0.
1=
The agent chooses x, h to maximize utility (x + h )
subject to px x + ph h = m +
h
h , where
px , ph are prices. The competitive demand function for health is
b
h = mb
ph
where pbh
3.1.2
ph +
is the private cost of health and
1
(px + pbh )
1
,
1.
Social equilibrium with a positive income externality from health
A planner with a utilitarian social welfare function would set h = h, so the budget constraint becomes
px x + ph h = m. The per capita social demand for health is then e
h = me
ph
is the social cost of health. Note that
1
(px + peh )
1
, where peh
ph
> 0 =) e
h>b
h. That is, the positive income externality of health
causes agents to consume less health than they would at the Pareto optimum.
5
3.2
Derived demand for health workers
Health is produced via a CES production function by two inputs, labor l and some other input z so that
h = h(z; l) = az z + al l
1=
1
, where "
choose z, l to maximize h(z; l)
pz z
1
is the elasticity of substitution and az , al > 0. Health systems
wl where pz is the price of z and w is the wage to labor in the health
sector.The derived per capita demand for labor in the health sector is
l(pz ; w; h) = a"l w
Note that
"
h.
(1)
> 0 =) l(e
h) > l(b
h), which is to say that fewer than the Pareto optimal number of health
workers are employed due to the health externality.
3.3
3.3.1
Do borders raise the number of health workers?
Destination country closed
There are two countries: the sending country (s) and the receiving country (r). With an impermeable border
between them, there are l0 health workers per capita— and L0 total health workers— in the sending country
who receive wage ws . Health workers in the receiving country earn wr .
3.3.2
Destination country open, inelastic labor supply
Suppose now that the border is eliminated so that some health workers in the sending country can sell their
labor in the receiving country at a new equilibrium wage w.
b Denote labor demand in the sending country
by ld , labor supply in the sending country by ls , and labor demand in the receiving country by l d . Labor
demand by the representative agent in the sending country is governed by ld = l(w;
b b
h), even though the
marginal social product is l(w;
b e
h). In the receiving country, health workers are paid their marginal social
product.
Migrants supply their labor abroad until the wage at home and abroad equalizes to w.
b Aggregate labor
demand in the sending country is Ld
P ld and in the receiving country is Ld
P ld , where P , P
are
populations. If L = Ld + Ld is the (…xed) total number of health workers in the two countries, and by
implication the labor supply curve in both countries is assumed vertical, then equilibrium number of health
workers in the sending country is
b=
L
L
1+
P e
h
P b
h
6
al
al
".
(2)
b is strictly increasing in b
Note that L
h, which is to say that underconsumption of health due to the externality
decreases the equilibrium number of workers who remain in the sending country. If wages prior to opening
b < L0 . Assuming that health workers in the
were higher in the receiving country, then unambiguously L
b
receiving country produce more health (al > al ), then @ L=@"
< 0. That is, the more that other inputs to
health production can substitute for health workers, the more of them depart.
3.3.3
Destination country open, elastic labor supply
Now relax the assumption that labor supply is inelastic in the sending country, and suppose that the opportunity to migrate is allocated randomly. As Stark et al. (1998) and Stark and Wang (2002) point out,
skilled workers could make the investment in health worker training based on the expected wage in the sector.
This is a weighted average of the sending country wage and the receiving country wage, according to the
Harris-Todaro (1970) condition we
wr + (1
)ws . The parameter , the fraction of all health workers
who emigrate, is de…ned by
b
L)=L
e,
(L0
where L0
(3)
b is the number of health workers who desire to emigrate, 0 6
L
6 1 is the fraction of those who
are permitted to do so, and Le is the equilibrium number of health workers in the sending country based on
the expected wage. In the equilibrium governing Le , workers’labor is demanded at wage ws but they supply
their labor according to the expected wage we such that
d
Lde
= a1 ws "
Lse
= a2 we" ,
(4)
s
(5)
where "s and "d are the absolute values of the wage elasticity of labor supply and demand. Substituting (4)
and the de…nition of we into (5) gives a single equation for Le and
with no simple closed-form solution for
either. But the implicit function theorem (after simpli…cation) yields
@Le
= Le "s
@
Note that as long as wr > ws , @Le =@
wr
wr + (1
ws
)(1 + "s ="d )ws
.
(6)
> 0. That is, if receiving country wages are higher than those in
the sending country, an increase in the fraction who emigrate raises the long-run equilibrium number of
health workers in the sending country. It does so by raising the expected wage at any given home wage, or
equivalently lowering the home wage that corresponds to a given expected wage— shifting the supply curve
down. If the receiving country allows no immigrants ( = 0) then (6) means that the percent decrease in the
7
equilibrium number of health workers in the sending country due to a one percentage point increase in the
number who depart is (@Le =@ )=Le =
wr
ws
1 =( "1s + "1d ). If this quantity exceeds unity, then starting from
autarky, an in…nitessimal increase in the number who depart produces a larger increase in the equilibrium
number who remain. Rearranging, this condition is
wr
ws
1>
1
1
+ .
"s
"d
(7)
In a way this restates the result of the preceding section, which showed that if labor is supplied inelastically,
increased emigration unambiguously lowers the equilibrium number of health workers in the sending country.
That is, (7) must be violated if "s = 0. But if wage gaps (wr =ws ) and labor supply elasticity are su¢ ciently
large, the condition (7) can always be met— the equilibrium quantity of health workers in the sending country
rises. Thus:
Proposition 1 If international wage gaps are su¢ ciently small and labor supply su¢ ciently inelastic, increased emigration must lower the short-run number of health workers in the sending country and the long-run
equilibrium number. Otherwise, increased emigration still lowers the short-run equilibrium number of health
workers in the sending country but the long-run equilibrium number can decrease, remain the same, or
increase.
3.4
3.4.1
Do borders raise welfare?
One sector
We have assumed throughout that there is a single sector in which health workers sell their labor in the
Rl
sending country. Per capita social surplus in the sending country is s1
(w(l)
e
ws ) dl where w(l)
e
0
= al e
h=l
1="
is the inverse social demand function for labor. Thus,
s1 =
When the health externality
is large then w
e
a"e
h " 1
l "
" 1
lws .
(8)
ws and s1 is large and positive, so that the departure of
the average health worker unambiguously lowers sending-country welfare. If condition (7) fails and b
l < l0
(openness to emigration lowers the number of health workers in the sending country) then social surplus in
the sending country unambiguously declines under openness. The larger the health externality, the greater
the decline (since @s1 =@ > 0).
An impermeable border thus raises the welfare of sending-country residents relative to even a partially
8
open border. If the social value of health is su¢ ciently large relative to the private value, the average
decline in sending-country welfare due to a departure (s1 =l) can exceed even the private gain to the migrant
(wr
ws ), meaning that an impermeable border raises the social surplus of sending country nationals as
a whole, not just sending-country residents. Again, by Proposition 1 this is contingent on a decline in the
equilibrium number of health workers under openness, which requires su¢ ciently small international wage
gaps and su¢ ciently inelastic labor supply in the sending countries.
3.4.2
Two sectors
Suppose now that the labor market for health professionals in the sending country is segmented into two
sectors. In the …rst sector, health workers are an input to the production of the public good h as before,
and private derived labor demand is l1 = aw
"b
h even though social demand is higher at e
l1 = aw
"e
h. In the
second sector, health workers are an input to the production of a private good h0 with the same price as h,
such that b
h < h0
is l2 = aw
e
h. Consumption of the good h0 has no positive externality. Derived demand in sector 2
" 0
h . Sector 1 might be thought of as the public system of primary health care. Sector 2 might
be thought of as the private primary care system or other nonclinical private employment.
R l1
Per capita social surplus in the sending country under autarky is now only s2
(w(l)
e
0
R l2 0
(w (l) ws ) dl, where w0 (l) a(h0 =l)1=" < w(l).
e
Solving gives
0
s2 =
where
" 1 2""
" a
1
ws1
"
" 1
2"
e
hb
h " + h0
1
"
lws ,
ws ) dl +
(9)
. From (8) and (9) follow three results: 1) s2 < s1 , thus the average social surplus
lost when the number of health workers declines by one is less when there is a second domestic sector. 2)
@s2 =@ < @s1 =@ , which is to say that when the externality is greater, the average social surplus lost due
to the loss of a health worker is lower when there are two sectors. Intuitively, this is because when the
externality is larger, workers in sector 1 are paid a smaller fraction of their marginal social product and more
choose to work in sector 2, where they generate a much smaller social surplus. 3) if " and e
h are su¢ ciently
high, or a and b
h are su¢ ciently low, then @s2 =@ < 0. Roughly speaking, if the externality is large and labor
demand is elastic, the average social surplus lost from the loss of a health worker declines. This arises from
the fact that a larger externality means that each health worker remaining in sector 1 has a larger impact on
public health, but fewer health workers supply their labor to sector 1. Under the above conditions, the latter
exceeds the former and each emigrating health worker carries a smaller social welfare loss as the externality
grows.
9
The addition of sector 2 has radically changed the model. When there is one sector, a larger health
externality means a greater loss of social surplus due to emigration. But when there are two sectors, all
else equal a larger health externality can mean that less social surplus is lost when any given health worker
emigrates. In essence the second sector is a ‘migration’ destination internal to the sending country with
greatly diminished social value but enhanced private value to the health worker. Domestic market failures
push health workers into this sector, and movement from this sector abroad entails only a small social loss.
Thus:
Proposition 2 If health workers in the sending country can sell their labor in a second domestic sector
producing a private good that conveys no positive externality, the loss of social surplus from the departure of
a representative worker 1) is smaller the larger is the externality of the public good, 2) might arise mostly
from lost production of a private good, and 3) is smallest in countries where the health externality is greatest.
This has an important implication. Earlier we saw that when there is a single sector in the sending
country and the elasticity of labor supply is su¢ ciently low, then imposing an impermeable border between
the sending country and the receiving country must raise welfare if the health externality is su¢ ciently large.
This reasoning falls apart, however, when either 1) the elasticity of labor supply is su¢ ciently large, since
removing the border can increase the equilbrium stock of health workers, or 2) health workers can produce
a private good in a second domestic sector, since the earlier welfare gain from imposing an impermeable
border is smaller as the externality in the …rst sector gets larger. It is easy to envision several extensions to
this model. For one, if health workers who emigrate are more likely to be selected from sector 2 than sector
1, the loss of social surplus from their departure is even smaller.
3.4.3
A graphical representation
This can be seen simply in Figure 1. The quantity of health workers in country s is measured from left
to right starting at Os ; the quantity in country r is measured from right to left starting at Or . AB is the
downward-sloping demand curve for labor in country r, where health workers are paid their marginal social
product. If health workers in country s are also paid their social product, labor demand in that country is
HA, and the equilibrium quantity of workers in r and s is Or L and Os L . If there is a border between the
two countries at L0 , opening the border raises world welfare by the area AF G.
But suppose health workers are paid less than their social product in country s (the “sending” country)
so that the competitive demand curve for labor there is BD. Opening the border still entails the gain BDF ,
but also the social loss BDGH; there is not only a loss to the sending country but a net loss to global
10
social welfare BF GH. Enforcing the border raises collective welfare, despite the harm to migrants and the
receiving country. A utilitarian social planner might support policy measures to prevent equilibration from
b such as restrictions on recruitment by country r of health professionals in s.
L0 to L
But two small re…nements might make us doubt this conclusion. First, the diagram has so far assumed
a vertical domestic supply curve in both countries. If in fact there is elastic supply of health workers in s,
following DI, opening the border could shift down this supply curve as workers are willing to accept lower
wages domestically in exchange for the possibility of migrating to earn the higher wage wr . The long-run
supply curve shifts out; the short-run curve shifts in. Whether openness increases or decreases the stock of
health workers at any given time— and thus whether it lowers sending-country welfare or global welfare— is
unclear.
Second, we have assumed so far that there is a single sector in country s. Suppose now that health
workers within s can choose to work either in sector 1, which produces a good with a positive externality, or
sector 2 which does not. The competitive demand curve for labor in sector 1 is JB, the social demand curve
for labor in sector 1 is KH, and the competitive (and social) demand curve for labor in sector 2 is F B. Note
that the larger is the externality, the lower is JB relative to KH, and the fewer workers choose sector 1.
Depending on the elasticities, the majority of health workers might be in sector 2 at any given moment, so
that the loss of social surplus due to any decline in the aggregate stock of health workers is correspondingly
smaller. It is smallest, in fact, in those countries where the health externality is largest. Health workers
“drain”out of sector 1 prior to emigration. The welfare e¤ect of international migration becomes small, and
more and more of that small e¤ect comes not from the health externality in sector 1 (BHKJ) but from
privately captured surplus in sector 2 (BEF ).
Naturally this simplistic, static analysis omits entirely the possibility that part of the welfare gain to
migrants (BCDE) could ‡ow back to the sending country in the form of remittances and repatriated savings.
It also omits more complex dynamic e¤ects whereby the international circulation of workers could a¤ect
domestic health workers’ productivity or the demand for health care, both of which could raise sendingcountry welfare.
Under the assumptions above, theory is ambiguous about the magnitudes and even the signs of the e¤ect
of relaxed migration policy ( ) on the stock of health workers in the sending country and social welfare there.
We can only reckon the overall e¤ect empirically. A quantitative answer requires three elements: a measure
of health professional emigration, a measure of sta¢ ng and health outcomes, and some exogenous variance
in emigration across di¤erent populations. The following three sections discuss each of these in turn.
11
4
An empirical approach
4.1
Regression model
What is the e¤ect of health worker departure from the sending country on the health worker stock of the
sending country? A simple regression model to render equation (6) is
ln b
li =
+ ln li0 +
i,
(10)
where b
li is the stock of health workers per capita in the sending country, and li0 is the cumulative number
of health workers who have left the receiving country and were health workers at the moment of departure.
Suppose we observe only li , the stock of health workers in the receiving country, per capita in the sending
country, such that li0 = ci li where ci is some country-speci…c constant. Plugging this into (10) gives
ln b
li =
+ ln li + f
An OLS estimate of (11) will give biased estimates of
f
i
i
+ ln ci g .
(11)
to the extent that li is correlated with the error term
+ ln ci g. We can reduce this bias by 1) controlling for other factors likely to cause both li and ci , such
as income and con‡ict, 2) seeking instruments for li that are uncorrelated with f
proxies for li0 besides li , and 4) observing whether inference on
i
+ ln ci g, 3) trying other
is sensitive to exclusion of observations
with extreme values of ci . The analysis below attempts all of these.
Now suppose that health care availability and health outcomes (both represented by hi ) in the sending
country are a function of health workers per capita in the sending country. A simple regression model to
render equations (8) and (9) is
h i = h0 b
li
0
i (li )
=) ln hi = ln h0 +
where h0 is some parameter common to all countries, and
ln b
li + ln
0
i (li )
0
i (li ),
(12)
captures some e¤ect of health worker
emigration on the quality of health workers in the sending country. Perhaps emigration siphons o¤ the most
capable health workers, for example. Alternatively, emigration of health workers from urban areas might
pull domestic health workers out of rural areas and into urban areas, diminishing the mass availability of
care. Let
i
(li0 )
Ai (li0 ) = Ai (ci li ) , thus
ln
i
= ln Ai + ln ci + ln li .
12
(13)
Plug (11) and (13) into (12) to get
ln hi = fln h0 +
g+(
+ ) ln li + fln Ai +
i
+(
+ ) ln ci g .
That is, the e¤ect of health worker departure on health is the sum of a direct e¤ect
(14)
, the degree to which
health worker departure lowers the number of health workers in the sending country (10) times the degree
to which fewer health workers damage health (12), and an indirect e¤ect
capturing e¤ects of departure
on the quality of those who remain in the sending country. A regression based on (14) will generate biased
+ to the degree that the size of the health-worker diaspora li is correlated with
estimates of the coe¢ cient
fln Ai +
i
cause li , Ai ,
+(
i,
+ ) ln ci g. This bias diminishes if we 1) control for other factors likely to simultaneously
and ci , 2) seek instruments for li uncorrelated with fln Ai +
proxies for li0 besides li , and 4) observe whether inference on
+
i
+(
+ ) ln ci g, 3) try other
is sensitive to exclusion of observations
with extreme values of ci .
Notably, lbi is not the right-hand side of (14). More health professionals at home are certainly posited to
increase health (through
i ).
But as health conditions can obviously have a reverse e¤ect on the number of
health professionals, this e¤ect is not identi…ed without proper instruments for lbi , which are not explored in
this study. To the extent that the instruments for li are valid, however, relegating lbi to the error term does
not bias the estimate of
4.2
+ .
Measuring African health worker emigration
The central reason for the lack of empirical research on the e¤ects of international migration is the absence
of a comprehensive database of bilateral international migration ‡ows. Even aggregate ‡ows between most
country pairs are unknown, to say nothing of age- or occupation-speci…c ‡ows. This study required building
a novel database of the cumulative net emigration of doctors and nurses from each African country. The
database simply counts how many African-born doctors and professional nurses were living and working in
each of the nine most important destination countries for skilled African emigrants at the time of the most
recent census in the destination country. The receiving countries comprise Australia, Belgium, Canada,
France, Portugal, South Africa, Spain, the United Kingdom, and the United States, whose last censuses
occurred between 1999 and 2001.
The resulting matrix of 53 sending countries and 9 receiving countries, presented by Clemens and Pettersson (2006), is the …rst occupation-speci…c database of skilled-worker net bilateral migration for a large
number of developing countries with a consistent de…nition of what constitutes an “emigrant”. These data,
13
combined with WHO estimates of the number of doctors and nurses who live and work in each African
country, allow easy estimation of the fraction of doctors and nurses from each country who lived and worked
abroad in the year 2000. Figure 2 presents these estimates for physicians; Figure 3 does the same for
professional nurses. Some considerations are necessary to properly interpret these estimates.
4.3
What is an ‘African’doctor?
There is no single ideal measure of “African health worker emigration”. One can interpret each component
of the phrase in multiple ways. Is an “African”someone born in Africa, someone whose ancestors for several
generations were born in Africa, someone trained in Africa, or someone who holds African citizenship? Does
“Africa” include North Africa and all of South Africa? Is a “health worker” someone who was trained as
such or someone who currently works in the health sector? How long must one stay outside the country to
be an “emigrant”?
This analysis classi…es “Africans” by country of birth; it includes the entire African continent; it counts
as doctors and nurses only those currently employed as doctors and nurses; and it calls “emigrants” those
who were residing in each receiving country on a su¢ ciently permanent basis circa 2000 to be included in
that country’s most recent census. These choices make the data suitable for the present purpose but less
useful for certain other purposes. Take the example of a physician born in Zimbabwe and trained in South
Africa who lives in Canada and holds on o¢ ce job in the pharmaceutical industry. Researchers interested
in the incentives faced by young educated people in Zimbabwe might be more interested in her country of
birth, while those interested in the …scal impact of emigration on South Africa might be more interested
in her country of training. Those interested in the relationship between migration and Black Economic
Empowerment might be more interested in her ethnicity than her country of birth. Those interested in the
speci…cally medical dimension of professional emigration might be interested to know that she was trained
as a clinician despite the fact that she does not now work as such.
Most studies that have attempted to measure African phyisician emigration previously have primarily or
exclusively classi…ed ‘emigrants’according to their country of training (Hagopian et al. 2004; Mullan 2005;
Bhargava and Docquier 2006; WHO 2006, p. 100). But this is not the single or even the most accurate
measure of professional emigration. This approach would lead to decent statistics for a study of, say, the
…scal consequences of physician emigration; the vast majority of African-trained doctors are trained with
public funds. (It would be problematic even for this purpose, however, since a portion of African doctors
trained abroad do so using scholarships funded by their home governments.) But a narrow focus on country
14
of training would not be appropriate for other studies— such as an investigation of the e¤ects of physician
emigration on health system sta¢ ng, health care availability, or health outcomes in the countries of origin.
To see this, note that 12 of the 53 countries in Africa (and 11 of 48 Sub-Saharan countries) do not have
a medical school accredited by the Foundation for Advancement of International Medical Education and
Research, or FAIMER (Educational Commission for Foreign Medical Graduates 2007). A medical degree
from a FAIMER-accredited school is a prerequisite to licensure in major receiving countries such as the
United States (American Medical Association 2007), and related but e¤ectively similar restrictions apply in
Australia and Canada. This means that, properly measured, an indicator of physician ‘drain’based strictly
on country of training would de…ne about a quarter of Sub-Saharan Africa to have lost zero physicians to
emigration. It is certain, however, that physicians would have left most or all of those countries to some
degree at some point, with possible consequences for sta¢ ng, the availability of care or health outcomes.
For related reasons, a country-of-training based measure would arti…cially de…ne nurse emigration from
most Francophone African countries to be extremely small, since French law currently mandates that only
graduates from French nursing schools may practice as professional nurses in France. Home-trained nurses
who leave must therefore train again, in France, in order to appear in the data as practicing nurses in France—
so they become foreign-trained nurses. Beyond this, a country-of-training measure for either doctors or nurses
would give an odd accounting even for countries that do have accredited schools but many of whose nationals
nevertheless train in other African countries. For instance: Doctors in the UK who were born in Zambia
and Zimbabwe, but who trained in South Africa, would contribute exclusively to the South African ‘brain
drain’— a classi…cation that might be sensible for some research questions, but not others.
Certainly a country-of-birth measure is problematic in that includes people who may have become health
professionals after emigration (and may not have become health professionals had they stayed). But a
measure based strictly on country of training is also highly problematic. The literature must make progress
on rigorous de…nition of what constitutes a skilled emigrant. This study simply uses both approaches and
observes that the de…nition does not a¤ect the …ndings here.
As this literature advances, we must avoid mixing the di¤erent classi…cations. In Bhargava and Docquier
(2006), African physicians abroad are counted in some destination countries by their country of birth, in
others by their country of citizenship, and in others by their country of training. This fact renders the
blended database extremely problematic. To give one example, the French Ordre National des Médecins
reports that in 1999 there were 238 physicians in France who were citizens of Sub-Saharan African countries,
but the French census of 1999 reports 4,203 physicians in France who were born in Sub-Saharan Africa— a
15
di¤erence of 1,766%! To make another comparison, in 2001 the Canadian Medical Association reports 190
physicians in Canada trained in Egypt, but the census of the same year shows 750 Egyptian-born physicians
(a 395% di¤erence). Such discrepancies are the rule, not the exception. Clearly, mixing these di¤erent
classi…cations can destroy the ability of the resulting number to measure anything at all. It is imperative to
choose a single de…nition of “emigrant” and retain it.
The case of Mozambique further illustrates the sensitivity of data like these to di¤erent assumptions. The
Mozambique Medical Association estimates that only around 5% of Mozambican physicians work abroad.
Destination-country census data show that about 75% of people born in Mozambique who now work as physicians do not live in Mozambique. The main cause of this disparity is the fact that many of those physicians
are of European ancestry and departed in the mass exodus of Portuguese colonists around independence in
1975. But it is not at all obvious that counting whites results in a poor measure of human capital loss. In
South Africa white health professionals today play an important role in educating a new generation of black
health professionals. It is true that Mozambican-born physicians in the white colonist class were providing
most of their health care to urban elites in the colonial era rather than to rural blacks, but the same could
be said of many black physicians in today’s independent African states.
4.4
Nine destination countries proxy for the world
This analysis also relies on the assumption that counting how many African health professionals live and
work in the nine most important destination countries gives an excellent proxy for how many live and work
in any foreign country. Those nine countries are the United Kingdom, United States, France, Australia,
Canada, Portugal, Belgium, Spain, and South Africa. I selected these seeking a balance between coverage
and the time and expense of additional data collection.
The primary reason that using these destination countries gives a good proxy for total health professional
emigration is that the …rst eight receiving countries alone account for 94.2% of all African-born, universityeducated people residing in any OECD country in 2000.1 My experience comparing the migration patterns
of African health professionals to those of other types of well-educated migrants suggests that the proportion
of total African health professional emigrants is similar to this value. I add a ninth country, South Africa,
because I take it to be the most important non-OECD receiving country for African health professionals.
Previous datasets that have focused on a single receiving country, such as that of Hagopian and others
(2004), do not give a full picture of emigration rates by sending country since bilateral ‡ows vary greatly
across sending countries.
1 Calculated
using the census data from Dumont and Lemaître (2005).
16
It is of course possible that another non-OECD country, such as Saudi Arabia, is important for some
countries, or that health professionals di¤er greatly in their migration patterns from other skilled professionals. But survey data from African health professionals considering emigration suggest that neither of
these is the case. Between 2001 and 2002, the World Health Organization interviewed 2,382 doctors, nurses,
and other health professionals in six African countries (Awases et al. 2004, p. 38). Each person declaring
an intention to emigrate was asked his or her favored destination. The fraction of these in each country
who gave one of our nine destination countries was 89.3% in Cameroon, 91.8% in Senegal, and 94.6% in
South Africa. A small percentage of respondents in Zimbabwe mentioned Botswana and New Zealand as
destinations but the vast majority mentioned one of our nine receiving countries. Respondents from Ghana
and Uganda did not mention any countries outside Africa besides the US and UK, and these two destinations
plus South Africa accounted for the vast majority of favored destinations in both cases.
Oosthuizen (2005, p. 177) surveyed in 2002 the favored destination countries of a sample of Registered
Nurses in South Africa who had just …nished their training if they were to work outside South Africa.
Of these, 24% mentioned countries outside Africa not included in the nine considered here: Ireland (2%),
New Zealand (4%), and Saudi Arabia (18%). An additional 11% mentioned unspeci…ed “other countries in
Europe and Africa”, a subset of which may be included in the nine countries considered here. These results
are somewhat di¢ cult to interpret since, of the 105 people who answered the survey, only 85 stated that
they would ever consider working outside the country while 91 gave a favored destination if they were to
work outside the country. The 105 respondents were self-selected from a pool of 500 inititally contacted,
so nonresponse bias in these numbers is a real possibility. Note also that direct recruitment of nurses by
Saudi Arabia in South Africa is a very recent phenomenon, meaning that the proportion of emigrating South
African Registered Nurses who went to Saudi Arabia before the year 2000 is certainly much lower than 18%.
Both in the surveys of Awases et al. and of Oosthuizen a small fraction of emigrating African health
professionals reveal the intent to work in another African country, a ‡ow which is not captured by the data
presented here and which represents a small discrepancy between these numbers and true emigration to
all other countries. It is smaller still when one considers reciprocal ‡ows: A small number of emigrating
Nigerian physicians go to work in Ghana, but a small number of emigrating Ghanaian physicians go to work
in Nigeria. Counting each as an additional loss would ignore the fact that for intra-Africa movements, one
country’s loss is another’s gain. And this discrepancy, to the extent that it is small and largely independent
of country characteristics, contributes primarily white noise to the data here rather than any bias that would
a¤ect the analysis. In sum, the true number of health professionals working abroad may exceed the number
working in the nine destination countries focused on here by an amount on the order of 5-10%.
17
4.5
Measuring outcomes
So much for the independent variable. Measuring the quantity of health workers per capita in di¤erent
African countries, on the left hand side of equations (6) and (11), is relatively simple. Decent statistics
can be had from the World Health Organization’s Global Health Atlas. Though surely including some
measurement error that is occasionally substantial, such error is in the dependent variable and probably not
so far from classical that it biases the coe¢ cient estimates.
More challenging is to measure “welfare”in equation (6) or “health”in equation (14). This study proxies
for these unobservables with a range of indicators of the mass availability of primary health care and of
basic public health conditions. Certainly several other indicators could be used, such as the availability of
specialized tertiary care or mortality due to a speci…c medical condition whose incidence is limited relative to
the greater population. The decision of which indicators to use was primarily a function of policy relevance;
policy discussions of medical migration focus on its e¤ects upon countries where generalized health conditions
are so dire, not on its e¤ects upon the availability of cancer treatment. An ancillary reason is that no
international organization publishes country-by-country data on the availability of tertiary care for conditions
like cancer, heart disease, or diabetes in Africa.
5
Simple correlations
5.1
Correlation of health worker stocks abroad and at home
The indicator of sta¢ ng impact used here is therefore the total number of physicians or professional nurses
in the country in the year 2000 as reported by the World Health Organization. Sta¢ ng e¤ects might well
be measured by other metrics, such as the degree of rationing in the health worker labor market (vacancy
rates) or shifts in the domestic geographic distribution of professionals. But the model makes clear that at
a minimum, if emigration constrains sta¢ ng in the health system as a whole it should decrease the total
number of health professionals in the country.
Figure 4 plots domestic physicians against physicians working abroad, each expressed per capita of the
sending country, for all 53 African countries. The simple correlation is unmistakably positive.
A correlation of this kind establishes little about the absolute magnitude of any particular pathway of
causation between the two variables. But it does contain a great deal of information about the relative
magnitudes of di¤erent causal forces at play on the continent. In Figure 2 we observe tremendous variation
18
in the tendency for physicians to leave di¤erent African countries, and in Figure 4 we are faced with the
simple fact that even these massive movements have been powerless to strip the most a¤ected countries
of their physicians. The negative impact on physician stock due to emigration, an accounting identity, is
obviously only one of several interacting causal forces determining this net positive correlation. There are only
three possibilities: Either greater domestic physician stocks must be driving greater emigration, or greater
emigration must somehow be encouraging the expansion of domestic physician stocks, or unobserved country
charactertistics must be simultaneously encouraging both expanded physician emigration and expanded
domestic physician stocks (or some combiation of these).
The next section will attempt to distinguish among those forces. But even before entering that discussion,
one thing is clear: Whichever of these three other forces is at work, it or they utterly overwhelm the
accounting-identity “drain” of physicians via emigration as a determinant of domestic physician sta¢ ng.
Put di¤erently, even the simple correlation of Figure 4 makes clear that if one is interested in knowing why
physician sta¢ ng is low across the entire health system of a particular African country, one learns little from
counting how many physicians have left that country. The primary causes of sta¢ ng constraints clearly lie
elsewhere. The below discussion of Table 3 makes clear that the emigration of professional nurses likewise
has a strong positive correlation with domestic nurse stocks; the same reasoning applies also to them.
5.2
Correlation of health worker stocks abroad with public health outcomes at
home
Likewise one can think of various measures of health care or health outcomes that might be a¤ected by
the departure of health professionals. Advanced tertiary care might be a¤ected, such as intensive care for
newborns or cardiac surgery, as might basic primary care for malaria and diarrhea. Since the migration
literature commonly refers to rich countries’ admission of emigrant health professionals as “poaching”—
which I de…ne as the appropriation of a public good for private gain— I focus here on public health impacts.
If physician and nurse emigration adversely a¤ects treatment options for the masses of low-income people
in African countries its e¤ects should be seen in broad public health indicators. These might include the
mortality rate under age …ve, the fraction of births attended by trained health personnel, and the fraction
of persons with HIV receiving modern treatment.
This is not to deny the importance of tertiary care. But the marginal utility of those receiving advanced
tertiary care in Africa, elites in their great majority, arguably does not di¤er greatly from the marginal
utility of tertiary care to the middle classes of rich countries. If there is to be a global welfare loss from
19
shifting health care from poor to rich countries, then, it is likely to be found in the basic primary care whose
marginal utility to low-income Africans is high.
Figure 5 plots the stock of physicians abroad against what is perhaps the fundamental indicator of health
conditions in the mass of low-income people: the mortality rate under age 5, per 1,000 live births. Children
do not appear to be markedly healthier in Niger, where very few physicians have left, than in Ghana, where
the majority have left.
Again, this says very little about the absolute magnitude of causation from physician emigration to
public health conditions among the poor. A small increase in child mortality due to physician emigration
is compatible with the simple negative correlation of Figure 5 provided there are other, confounding forces.
But it does suggest that any such e¤ect is utterly swamped by the other pathways of causation. Once
again, there are only three possible confounding e¤ects: Either lower child mortality somehow causes greater
physician emigration, or greater emigration must somehow cause lower child mortality, or unobserved country
charactertistics must be simultaneously encouraging both expanded physician emigration and lower child
mortality (or some combiation of these). The section to follow explores this issue. But without going any
further, Figure 5 makes it clear that if we want to know why low-income children are dying in a particular
African country, we learn nothing from counting how many physicians have emigrated. The discussion of
Table 5, to follow, demonstrates that all of these arguments apply equally to professional nurses.
6
Measuring the e¤ects with natural quasi-experiments
Proposition 1 suggests that an exogenous change in openness to emigration can either raise or lower the
equilibrium domestic number of health professionals, depending among other things on the elasticity of labor
supply. Proposition 2 implies that the same change can raise welfare if the number of health professionals
rises, and if the number falls, openness can lower welfare to an arbitrarily small degree depending on the
structure of the domestic labor market. In short, the sta¢ ng and welfare e¤ects of an exogenous change in
openness to emigration are empirical questions.
Simple correlations show that if there is any negative, causal e¤ect of emigration on health and sta¢ ng,
it is small relative to the other forces shaping these variables. But we might also be interested in learning the
magnitude of the causal e¤ect, and the right natural experiment could help isolate and measure it. Another
reason to seek an exogenous driver of migration is that it will correct for attenuation bias due to measurement
error in the emigration variable. There is little reason to expect that physician counts by country of birth are
20
measured with any substantial error by receiving-country governments, but perhaps measuring emigration
by country of birth is a measurement with error of some other emigration indicator that most a¤ects health
and sta¢ ng.
The simple model above directly yields two candidates for plausibly exogenous instruments. Note that
equations (2) and (3) imply that @ =@
> 0 and @ =@(P =P ) > 0. In other words:
Proposition 3 When immigration policy is more open in the receiving country a larger fraction of the
sending-country health workforce departs. Likewise, when the receiving country is large relative to the sending
country, a larger fraction of the sending-country health workforce departs.
All else equal, then, sending-country size should negatively a¤ect emigration; receiving-country size
should positively a¤ect emigration. There is little reason to think that either has a substantial direct e¤ect
on health. Sending-country size has been used as an instrument for emigration by Beine et al. (2003). But
African history allows us to gain further identi…cation from the fact that some emigrants are exogenously
channeled to some receiving countries.
6.1
Two natural experiments arising from the colonial division of the continent
European colonial interventions in the 18th and 19th centuries largely …xed the modern international frontiers
of Africa. But di¤erent European powers did not use markedly di¤erent criteria in selecting the locations of
initial outpost settlements. And when the interior of the continent was carved up in a series of conferences,
disputes, and treaties in the years leading up to 1900, “boundaries were completely arti…cial: heterogeneous
tribes were lumped together, while watersheds and meridians cut some tribes asunder”(McEwan 1968, 214).
Whether any given doctor in the middle Niger River valley grew up in tiny Benin speaking French or in giant
Nigeria speaking English has little to do with her personal traits or with local health conditions. But there
are several reasons to believe that her country a¤ects her propensity to emigrate. The two most important
are the language of training (which a¤ects professional opportunities abroad) and the size of her country
(which a¤ects professional opportunities at home).
I argue here that neither of these is likely to directly a¤ect health system sta¢ ng levels, basic primary
health care, or basic public health outcomes. To the extent that this is true, language and country size create
exogenous variance in emigration rates, and are valid instruments for emigration rates in the regressions that
follow. Said di¤erently, variance in African countries’dominant languages and sizes constitute two natural
quasi-experiments permitting estimation of the causal e¤ect of emigration on sta¢ ng, basic care, and public
health.
21
6.1.1
Linguistic heritage
For reasons having little to do with their economic conditions, geography, culture, or skill, a large fraction
of African health professionals are trained in French and a di¤erent, large fraction is trained in English.
If this historical accident substantially a¤ects emigration rates, then, francophonie might serve as a valid
instrument in regressions seeking the determinants of health professional sta¢ ng and health outcomes.
Assume for a moment that the ability of any given African health professional to obtain a job in one of
the principal destination countries is a function solely of the availability of positions. An African physician
considering employment abroad could join a labor market of approximately 210,000 physician jobs in France
and Francophone Canada, or a market of about 860,000 physician jobs in the United States, United Kingdom,
Australia, and Anglophone Canada. Assume further that the fraction of positions that are available at any
given time is roughly similar across destinations. Since a major part of the immigration barrier in all of these
countries is a rigorous language requirement, this suggests that all else equal, an English-speaking African
emigrant physician has about four times the job opportunities abroad. For professional nurses, the major
Francophone destinations hold about 470,000 jobs whereas the major Anglophone destinations hold about
3.2 million. English-speaking African professional nurses, then, have roughly six to seven times the number
of job opportunities abroad.
But of course the availability of positions is not the only barrier. Relaxing this assumption we see that the
di¤erence in opportunities abroad is even more pronounced between Francophone and Anglophone countries.
In addition to the language requirement, several other barriers vary across destinations: Notably licensing
requirements are somewhat more relaxed in Anglophone than in Francophone countries, for both doctors and
nurses. Newly-arrived physicians trained in Africa who wish to practice in France must repeat the process
of medical residency, practicing under supervision for three years, before becoming licensed for independent
practice. Few exceptions are granted. In the United Kingdom, in contrast, somewhere between three months
and three years of supervised practice are required, decided ad hoc by General Medical Council. In Australia
between one and three years are required— decided on an individual basis and varying across states— but
this requirement can be waived for those willing to practice in rural “areas of need”. In Canada, general
practitioners are required to practice under supervision for two or three years, though again waivers are
granted by some provinces for those practicing in “under serviced” rural areas. The United States requires
three years of supervised practice with few exceptions, but “J1” training visa policy makes it easier for
physicians to stay in the country for work following residency for those who will work in rural “Medically
Underserved Areas”.
22
The general picture that emerges is that France’s licensure requirements for foreign physicians are somewhat more stringent than those of the other major destination countries for African physicians, both de jure
and de facto. This may contribute to the fact that, if one omits exceptional Algeria, Morocco, and Tunisia
from the sample, African-born physicians make up about 2.4% of the physician workforce in France but 3.9%
of the workforce in the United Kingdom, United States, Australia and Canada collectively.
For nurses there is an even clearer di¤erence in licensure requirements. African-trained nurses wishing to
practice in France must …rst, regardless of their current quali…cations and experience, enroll in and graduate
from a French nursing school to earn the diplôme d’état. This is in sharp contrast with the major Anglophone
destination countries (and Francophone Canada), for which it su¢ cies to pass a language and skills test and
provide a transcript of prior training credential. This di¤erential barrier only substantially a¤ects those who
are foreign-trained (who constitute, in the US for example, a minority of all African-born professional nurses).
But it is likely one reason why— again omitting exceptional Algeria, Morocco, and Tunisia— African-born
professional nurses make up about 1.1% of the professional nurse workforce in France but 1.4% in the major
Anglophone destinations collectively.
Not only, then, are there more opportunties abroad for English-speaking African health professionals,
but those opportunities are more easily accessible. One might expect, then, that emigration rates for both
physicians and nurses are substantially higher in Anglophone than Francophone countries, and below we
will see that this is indeed the case. But francophonie is not a valid instrument in regressions determining
sta¢ ng or health outcomes unless it has no e¤ect on these through other channels than migration.
Colonial a¢ liation might directly a¤ect sta¢ ng and health in two ways: one arising from where di¤erent
colonial powers chose to settle, another from how they ran their colonies. Di¤erent colonial powers might
have chosen which parts of Africa to colonize according to local traits likely to a¤ect today’s sta¢ ng and
health. They might also have run their colonies in ways whose di¤erences generated systematic di¤erences
in today’s sta¢ ng and health.
There is little evidence that colonists’ initial coastal settlements or the subsequent partition of the interior took place according to criteria of development potential, natives’characteristics, or reigning disease
conditions. Historians are clear that at the time of settlement and partition, “there was little or no exact
knowledge of the resources available in Africa or of the problems which might be involved in developing
them pro…tably” (Fage 1978, 321). “Statesmen and diplomats met in o¢ ces or country houses and drew
lines across maps which themselves were usually inaccurate”(Oliver and Fage 1963, 189). “The …nal division
of territories re‡ected not so much the strength of European interests on the ground as the political power
23
of the claimants in Europe” (Oliver and Atmore 2005, 129). After partition, the European powers “had
few speci…c plans about what to do with them. Much had been acquired simply for the purpose of keeping
other powers out. Speci…c possibilities suggested the development of gold mining here, cotton growing there
... but long-range plans for economic, social, or political change were very few indeed” (Curtin et al. 1978,
472-3).
To be sure, each European power was motivated partly by the economic potential of the colonies—
alongside more prominent motives of national prestige, military strategy, evangelism, and diplomatic advantage (Sanderson 1985, 116; Gann 1969, 113-119). And borders were sometimes drawn with regard to intact
African polities; the border between Nigeria and French West Africa was set partly in accordance with preexisting tribal territories, as was that between Angola and Rhodesia (Touval 1966, 289). But these motives
did not di¤er systematically accross di¤erent European powers. Sanderson (1985, 105-6) ‡atly concludes
that “no economic model can explain, even in the broadest and most general terms, the pattern of partition
that actually took place.”
It is nevertheless true that, simply by chance rather than design, French-speaking settlers may have ended
up in regions that are naturally more prone to disease. Many Francophone countries cluster close to the
equator not far from the Gulf of Guinea, where tropical diseases are more prevalent than at other latitudes.
Public health outcomes may then be materially di¤erent in Francophone countries— though this would not
a¤ect regressions whose dependent variable measures treatment e¤ort or domestic supply of professionals.
Empirically, however, the non-Francophone countries are almost as tropical as the Francophone. A simple
cross-country average of the fraction population living in the tropics, for example, is 0.85 for Francophone
Africa and 0.80 for non-Francophone (Gallup et al. 1998).
The other way in which colonial a¢ liation could directly a¤ect sta¢ ng or health is through di¤erent
powers’ di¤ering treatment of the colonies after partition. La Porta et al. (1997) show that countries
with French civil law have weaker investor protections than others and less-developed capital markets.
Francophone countries may thereby have poorer general economic conditions and more di¢ cult business
conditions for health professionals. Furthermore, Francophone and Anglophone educational systems di¤er
substantially, in ways that could theoretically a¤ect the quantity or quality of health care professionals, with
possible consequences for treatment and health outcomes as well.
Empirically, however, these theoretical concerns do not provide a strong case for nonexcludability of the
Francophone instrument in the regressions to follow. First, the p-value of a t-test of the null hypothesis that
mean income per capita is equal between Francophone (mean 714) and non-Francophone countries (mean
24
1143) is 0.28, failing to reject the null. The non-Francophone mean falls to 934 and the p-value is 0.46
if one excludes the Seychelles from the sample. Nevertheless, to be on the safe side, all regressions using
Francophone as an instrument include income per capita to account for systematic di¤erences in general
economic conditions between Francophone and non-Francophone countries. Second, achievements of higher
education in the two groups of countries do not greatly di¤er. The fraction of the population over age 15
who has completed tertiary education in Francophone African countries (1.1%) is very similar to the fraction
in non-Francophone (1.0%) (Barro and Lee 2000). To be conservative here, we nevertheless include a basic
measure of the strength of the educational system— gross primary school enrollment— to all regressions that
use francophonie as an instrument.
6.1.2
Country size
It is a well-known theoretical and empirical result that smaller countries are more open to trade (e.g. Alesina
et al. 2005; Rose 2006). Country size has been exploited as a valid instrument for trade (e.g. Frankel and
Romer 1999). The same models with insubstantial changes predict that smaller countries will be more
open to international movements of people as well— for the same reasons of economies of scale in input and
output markets and in complementary goods and services. Docquier and Rapoport (2004) document that
net emigration rates per capita in the sending country do decrease monotonically with sending country size,
so that aggregate emigration rates are less than a …fth in countries with population over 25 million of what
they are in countries under 4 million. Docquier and Marfouk (2005, p. 19) note the empirical regularity
that emigration rates are higher, all else equal, in smaller countries, and that “these di¤erences cannot be
attributed to the educational structure of residents or to a stronger selection in migration ‡ows. Smaller
countries simply tend to be more opened to migration.”
This should hold a fortiori for skilled worker migration, since one might expect greater economies of
scale in production inputs complementary to high-skill workers than low-skill. Intuitively, imagine taking
any single country on earth and randomly chopping it into multiple smaller countries of di¤erent sizes. Since
economies of scale are greater for hospitals than for primary schools, this would result in greater rates of labor
“migration”across the new borders by hospital workers than by primary school teachers. The largest of the
new mini-countries would be much more likely to contain more hospitals per capita than the smallest minicountries, but not substantially more likely to have more primary schools per capita. Labor “emigration”
from the largest mini-countries would thus be lower among relatively high-skill surgeons who work at the
hospitals than among relatively low-skill primary school teachers, simply because hospitals must be few and
large to work properly.
25
These complementary goods and services are more available in larger African countries. A good proxy
for the availability of complementary goods and services for health professionals in a country is the existence
of an internationally accredited medical school in that country. Sustained operation of an internationalclass medical school requires myriad goods and services complementary to the profession. Hagopian et al.
(2004) catalog the number of medical schools in each African country accredited by the Foundation for the
Advancement of International Medical Education and Research (FAIMER). All 11 African countries with no
FAIMER-accredited medical school have population under 12 million and all but one have population under
4.5 million. Of the 36 countries with at least one FAIMER-accredited school, all but 7 have a population
over 5 million.
For country size to serve as a valid instrument for health professional emigration, however, it must also be
excludable from a regression with health sta¢ ng, primary care, or basic health outcomes as the dependent
variable. In this case, that means that country size should not a¤ect any of these variables except through
its e¤ect on emigration rates. There is a strong case for excludability. Recall that the emigration indicator
used here is based on country of birth. Nothing about the argument above suggests that a person born in a
smaller country is more or less likely to be a physician, all else equal, than someone born in a larger country.
That person may choose not to work as a physician in a place with few goods and services that complement
the profession (small countries), but this e¤ect acts solely through migration. It may be, of course, that
for other reasons than those discussed so far, being small makes a country less likely to raise children who
choose to become or are able to become health professionals. Small countries could be poorer, for example,
and poverty could stunt their mental development. Small countries could also experience greater emigration
of teachers, limiting children’s access to childhood learning that enables them to qualify for professional
schools.
But there is little empirical support for this theoretical concern. Rose (2006) tests whether or not smaller
countries di¤er from larger ones on a wide range of development indicators. He …nds no systematic di¤erences
due to country size per se in income per capita, public health outcome and treatment indicators like child
mortality or vaccination rates, and indicators of the strength of the education system such as the literacy or
the primary school completion rate.2 This suggests that there is little reason to be wary of the excludability
of country size in this case. This empirical evidence aside, there is much theoretical unease in the trade
literature on the validity of geographical variables like country size in regressions like this, and Rodríguez
and Rodrik (2001) provide evidence that some of Frankel and Romer’s geographic instruments for trade in
2 Rose actually uses ln(population) as his measure of size rather than land area, but in the 53 African countries, the correlation
between ln(population) and ln(land area) is 0.79 (from the World Bank Africa Database 2004, data year 2000).
26
regressions with income per capita as the dependent variable are not excludable (though note that country
size is not among these). To be on the safe side, we thus include in the main equation income per capita
and basic schooling rates in all regressions that use country size as an instrument.
6.2
Other instruments
A further consequence of Africa’s international borders is that economic conditions in the natural receiving
countries vary importantly across sending countries. Doctors have high salaries in the United States, for
example, but this labor market is less accessible to African physicians who learned their trade in French.
This provides another potentially useful quasi-experiment.
A large body of empirical evidence suggests that relatively favorable economic conditions in the principal
destination countries are a powerful cause of emigration from Africa (e.g. Hatton and Williamson 2003).
To the extent that health system sta¢ ng and public health conditions in African countries are otherwise
una¤ected by destination-country prosperity, indicators of the latter could constitute valid instruments for
emigration. Here we use receiving-country income per capita and average profession-speci…c wages in the
receiving country as two such instruments.
This is made slightly complex by the fact that for several African sending countries, multiple destination
countries are important. Here we create for each sending country an index of destination-country income per
capita as the weighted average of income per capita across the nine principal receiving countries. A second
index is the weighted average of profession-speci…c wages in the nine principal receiving countries, gathered
by an international salary consulting …rm. When these indices instrument for physician (professional nurse)
emigration, the weights in each case are the fraction of physicians (professional nurses) born in that sending
country who live in each receiving country.
While it is di¢ cult to imagine how an increase in (say) physician wages in Belgium could per se directly
a¤ect public health in Burundi through any other channel than migration, it is much more likely that income
per capita in Belgium could a¤ect economic and social development in Burundi through multiple channels.
In such a case, the instrument would be nonexcludable from the main regression equation and thus invalid.
For this reason we run each instrumented regression with and without these two additional instruments, and
include always an indicator of income per capita in the main equation.
With these elements in place— an indicator of emigration, a set of outcomes of interest, and a set of
natural experiments— the data are in place to test the null hypothesis of no e¤ect of health professional
27
emigration on sta¢ ng and health outcomes. Table 1 summarizes all the variables, both unlogged and logged
(since all regressions are in logs).
6.3
Results: The e¤ect of health workers abroad on domestic stocks
Table 2 investigates the causal e¤ect of physicians abroad on the number of physicians in each African
country, both measured per capita in the sending country. All regressions are based on equation (11). The
unit of observation is the African sending country. The indicator of physician emigration in the …rst row
is the number of physicians who live and work abroad, in the year 2000, born in each sending country, per
capita in the sending country. Regression (1) is the simple bivariate correlation from Figure 4, revealing a
strong positive correlation.
What produces the positive correlation? Some of the most obvious candidates that could simultaneously
a¤ect both variables in an African context are wealth, education systems, and warfare. Regression (2)
controls for income per capita— strongly positively correlated with physicians per capita— as well as primary
school enrollment and a dummy variable for a major war between 1970 and 1999. The correlation between
physician emigration and domestic physician stocks weakens somewhat in magnitude by remains statistically
signi…cantly positive at less than the 1% level.
Additional omitted variables could still be driving this correlation, biasing upward the coe¢ cient on
emigration by generating correlation between that variable and the error term. We can address that violation
of the regression model’s assumptions with good instruments— correlated with emigration but with no direct
e¤ect on domestic physician stocks. Regression (3) shows that francophonie and country size are indeed
correlated with emigration to a statistically signi…cant degree. An F -test rejects the hypothesis that their
coe¢ cients are jointly zero at less than the 1% level. They jointly explain roughly one …fth of the variance
in physicians abroad per capita. Including desination physician wages or destination income per capita in
regression (4) does not improve identi…cation.
Regression (5) re-runs regression (2), this time instrumenting for physicians abroad per capita using only
francophonie and country size, the instruments that emerge from Proposition 3. Regression (6) uses all
four instruments from regression.(4). The sign on physicians abroad remains positive but the coe¢ cient is
statistically indistinguishable from zero. To the extent that the instruments are valid, there is no discernible
negative e¤ect of physician emigration on domestic physician stocks. Regressions (7) and (8) show the
…rst-stage results from regressions (5) and (6).
28
Regressions (9) and (10) employ the more e¢ cient three-stage least squares estimator to take advantage
of information contained in the cross-equation correlations of error terms. The coe¢ cient on the physician
diaspora size rises slightly and the standard error falls slightly.
Finally, regression (11) uses the stock of physicians outside each African country, per capita in the sending
country, calculated by Bhargava and Docquier (2006). While this measure combines country of training,
country of birth, and country of citizenship to de…ne an ‘African’physician abroad— di¤ering according to
the receiving country where data were gathered— it focuses on country of training in key destination countries
like the US, UK and France. It is thus an independent indicator of cumulative physician emigration using a
substantially di¤erent de…nition of ‘emigrant doctor’. The results do not change The correlation between
this measure of emigrant physicians and that in row 1 of the table, based on country of birth, is 0.73.
Table 3 repeats precisely the same analysis described above, with the sole di¤erence that professional
nurses rather than physicians are considered in every instance. The conclusions are the same in all but
insubstantial details.
6.4
Results: The e¤ect on basic health care and public health outcomes
Table 4 makes only one adjustment to the preceding analysis. The dependent variable in the upper half of
the table— regressions (1), (2), (5), (6), and (9) to (11)— is mortality in children under age …ve per 1,000
live births. All regressions are based on equation (14). The table is an analytical extension of the simple
correlation in Figure 5.
If physicians abroad at the massive levels seen in Figure 2 substantially degrades basic public health
conditions, we might expect a positive coe¢ cient on the migration variable. The coe¢ cient on physicians
abroad never positive and is often statistically distinguishable from zero. This remains true when two-stage
least squares and then three-stage least squares attempt to isolate the causal impact of physicians abroad
on child mortality, with increasing e¢ ciency. With the expanded instrument set that includes destination
physician wage and destination income per capita in regression (6), the instruments come close to failing
Hansen’s excludability test at the 5% level. Restricting the instrument set to the more parsimonious pair of
Francophone and land area, however, eliminates this correlation in regression (5). And these two instruments
are preferred at any rate, since the …rst-stage regressions (7) and (8) and the Anderson canonical correlation
test in regressions (5) and (6) make it clear that those two contain all the …rst-stage explanatory power.
The Anderson-Rubin test in regression (8) comes close to rejecting the hypothesis that the coe¢ cient on
physician emigration is zero, but note that the sign of the coe¢ cient is negative.
29
Table 5 repeats the above analysis for professional nurses. Again, there is no evidence that even the
huge numbers of nurses abroad seen in Figure 3 have any positive causal e¤ect on child mortality. The
qualitative conclusions from the preceding paragraph apply equally to this table, with two exceptions. One
exception is that Hansen’s test comes nowhere near rejecting the null of nonexcludable instruments in this
case. The other is that the hypothesis of a zero coe¢ cient on nurses abroad in all regressions is rejected at
the 5% level with the simple t or z statistic, and rejected in regressions (5) and (6) at the 10% level with the
Anderson-Rubin statistic. That is, these regressions are suggestive of a small negative causal e¤ect of nurse
emigration— or something it proxies for— on child mortality.
Child mortality is certainly not the only indicator of public health. Tables 6 and 7 repeat the threestage least squares results of Table 4, regression (9), for ten other indicators of basic primary health care
and public health outcomes that are widely available across African countries. These are: mortality under
age one per 1,000 live births (IMR); the measles vaccination rate; the diptheria/pertussis/tetanus (DPT)
vaccination rate; the prevalence of acute respiratory infections (ARI) in children under age …ve during two
weeks requiring medical care; the fraction of those ARIs that saw a trained health professional; the percentage
of deliveries attended by trained personnel; the percentage of children under age …ve with diarrhea requiring
medical attention over a two-week period, the percentage of those diarrhea cases who received either oral
rehydration therapy or increased ‡uids, with continued feeding; the percentage of people ages 15-49 who
are infected with HIV; and the percentage of adults with advanced HIV infection receiving antiretroviral
treatment.
The tests fail to detect any deleterious e¤ect of physicians abroad on basic health care availability or mass
public health outcomes in any of the ten cases. The coe¢ cients on all the treatment variables (vaccinations,
attended births, ARI treatment, diarrhea treatment, and HIV treatment) are uniformly positive. Thus
even more e¢ cient inference could not by itself reveal a negative e¤ect of physician emigration on basic
primary health care provision in these data. The coe¢ cients on public health outcomes are negative (for
infant mortality ARI prevalence, and diarrhea prevalence) with the exception of HIV prevalence— which is
statistically indistinguishable from zero.
Two additional tables that repeat the analysis of Tables 6 and 7 for professional nurses are omitted here
for brevity. There are two qualitative di¤erences between those tables and the discussion above. The …rst
is in the signs of a few coe¢ cients: The coe¢ cient on nurse migration is positive in the diarrhea prevalence
regression, though statistically indistinguishable from zero with a z-statistic of 0.09. The second di¤erence
is that the negative coe¢ cient on nurse migration in the regressions for infant mortality and ARI prevalence
30
are negative and statistically signi…cant, and the coe¢ cient in regression for HIV treatment is positive and
statistically signi…cant.
In sum, we observe no causal connection whatsoever between the very large cross-country variance in
African health worker diasporas and any of degradation of basic primary health care provision or public
health outcomes in the sending-countries.
6.5
Econometric issues
We might have several concerns in interpreting these results, at least including weak instruments, nonexcludable instruments, ine¢ cient inference, in‡uential observations, and di¤erent classes of measurement error.
6.5.1
Weak instruments
First, regressions (3) and (4) in Tables 2 through 5 show that the instruments do explain a substantial
portion of the variance in emigration and do so to a highly statistically signi…cant degree in terms of the
F-test. Likewise, the canonical correlation likelihood ratio test of Anderson (1951) rejects the null of weak
instruments at the less than 1% level in all cases. Hall et al. (1996) and Stock et al. (2002), among others,
argue that this is insu¢ cient to rule out weak instrumentation especially in …nite samples with invalid,
endogenous instruments. Stock et al. recommend as one solution the test of Anderson and Rubin (1949) of
the null hypothesis that the coe¢ cient on the endogenous variable in the main equation is zero, a test that
is fully robust to weak instruments. Here the null is not rejected at the 5% level anywhere except regression
(6) in Table 3, where the coe¢ cient on nurses abroad is positive.
Stock and Yogo (2005) calculate critical values below which the value of the Cragg-Donald (1993) statistic
would suggest bias in the coe¢ cient estimates. In the coe¢ cient estimate in regression (5) of Table 2, Stock
and Yogo’s critical values suggest that with 95% con…dence, the bias in the coe¢ cient estimates on physicians
abroad is less than 20% of the OLS bias.3 Instrumentation in regression (5) of Tables 3, 4, and 5 is likewise
‘strong’by the de…nition of Stock and Yogo.
6.5.2
Overidenti…cation
We might also be concerned that the instruments are endogenous and inappropriately excluded from the
main equation. Controlling for income per capita, schooling, and warfare in the main equation— super‡uous
3 Taking the coe¢ cients from columns (2) and (5), and letting x represent the lower bound on the true coe¢ cient, we have
0:20 (0:36 x) + x = 0:03, thus x = 0:05. That is, we can be 95% con…dent that a one-percent increase in the size of the
physician diaspora causes less than a one twentieth of one percent decrease in the stock of physicians in the sending country,
or an increase in that stock.
31
if the instruments are valid— allays this concern since the channels of direct causation from the instruments
to physician stocks would have to bypass those variables to invalidate the instruments.
Furthermore, we discussed above several reasons to expect a priori that the instruments are in fact
excludable. And the Hansen (1982) J-test fails to reject the null of zero correlation between the instruments
and the estimated disturbances at the 5% level in Tables 2 to 5. Since this test is based on estimated rather
than true residuals, and thus relies on the assumption that at least one of the instruments is in fact valid, it
is indicative rather than dispositive. But it does suggest that if there is any direct e¤ect of the instrumental
variables on physician stocks, it is very small. If tremendous cross-country di¤erences in ci from equation (11)
were driving these results— that is, if countries where health conditions were the worst were systematically
those countries where the largest fraction of the health worker diaspora became health professionals after
leaving— then we would expect to see a substantial correlation between the instruments and the error term,
which includes ci .
6.5.3
E¢ ciency
Next, we might be concerned that the limited information estimators in regressions (1) through (6) of Tables
2 to 5 fail to reject the hypothesis that the coe¢ cient on physicians abroad is zero because they fail to take
advantage of the additional information contained in the correlation between the error terms in the system
of two equations. Certainly we should expect that some of the same unobserved country characteristics that
create more doctors could also create more doctors who emigrate, information which a system estimator
could exploit. Note however that the coe¢ cient on health workers abroad is positive when the dependent
variable is domestic health worker stocks, and negative when the dependent variable is child mortality.
So ine¢ cient inference could not be a reason why the estimation does not detect any negative impact of
emigration on domestic health worker stocks or any positive impact on child mortality. As we move to a
full-information three-stage least squares estimator in regressions (9) and (10) of each table, we see no change
in these patterns.
6.5.4
In‡uential observations
In this small-sample setting we might be concerned that a single in‡uential observation is unduly a¤ecting
the coe¢ cient estimate on physicians abroad. Figures 4 and 5 certainly suggest that this is not the case for
physicians abroad or for domestic physician stocks per se. Among the covariates, an obvious candidate for
extreme distortion is the GDP per capita of Equatorial Guinea, which shot to the stratosphere after oil wells
came online during the 1990s and which grew 67% in 1997 alone. The results are not qualitatively sensitive
32
to the omission of any one country, including Equatorial Guinea. To the extent that the instruments are
valid, of course, the covariates should not a¤ect the coe¢ cient estimate on physicians abroad at all.
6.5.5
Nonclassical measurement error in health workers abroad
Not all health professionals living outside their country of birth would have become health professionals if
they had not left that country. To the extent that this fraction does not vary across countries it does not
bias the above estimates at all. To the extent that it varies across countries but is uncorrelated to the size of
the diaspora there is no bias either. But if it is correlated with the diaspora size, bias could be present. In
other words, if ci is strongly correlated with li , then the coe¢ cient estimates on li in regression equations
(11) and (14) will be biased.
Empirically, the best way to address this concern would be to measure li0 directly. As discussed above,
a country-of-training based emigration measure is a very imperfect measure of li0 , but is biased in di¤erent
ways than a measure based on country of birth. In Tables 2 and 4, using a measure that focuses primarily
on country of training does not change the result. Since all those trained in the country of origin were
health professionals at departure, this is a strong reason to believe that the other regression results are not
spuriously generated by correlation between li and ci .
Another approach would be to …nd an experiment provoking variance in li but not in ci , that is, exogenous variation in health worker emigration that does not change the incentives of people to become health
professionals if they could not leave the country. The natural quasi-experiments above may not meet this
criterion. The approach we take here is to …nd a variable likely to be correlated with the propensity to become a health professional in the counterfactual case of staying at home— age at emigration— and consider
whether or not it is correlated with total emigration li .
Certainly the age at migration must be strongly correlated with the propensity to become a health
professional in the counterfactual case of having remained in the country of origin. A person born in Nigeria
who works in the UK as a physician, and emigrated at age 45, is extremely likely to have been a physician
in Nigeria since extremely few foreign students who …nish a British medical degree do so after age 50.
Conversely, only a limited fraction of Nigerian-born physicians in the UK who emigrated at age 5 would
have been physicians if they had not left Nigeria. If those African countries with the largest health worker
diasporas also contain the largest fractions of those who departed at a young age, before becoming health
professionals, this would suggest that corr(li ; ci ) < 0 and that the estimates of the preceding tables re‡ect a
lower bound on the true e¤ect.
33
But we do not observe such a correlation in survey data on African-born physicians in the United States,
Canada, and France. Table 8 reports these results. The …rst row uses the Public Use Microdata Sample (5%)
of the United States census from the year 2000, which reports interviews with 614 physicians born in one of
18 African countries. Each individual reports an age in 2000 as well as the year of arrival in the US, allowing
unbiased estimation of that person’s age at arrival. This age is then averaged within each of the 18 countries
of birth. OLS regressions in columns 1 and 3 show that this average is not correlated across sending countries
to a statistically signi…cant degree with physicians per capita abroad or the fraction of physicians abroad,
the regressors from Tables 2, 4, 6, and 7. The reliability of these averages varies, of course: The sample
represents a population of 21 doctors born in Senegal but 2,767 born in Nigeria. For this reason columns 2
and 4 repeat the regressions with analytical weights inversely proportional to census population estimates
for each country of birth. There is no meaningful correlation between the age at emigration and physicians
per capita abroad. There is weak evidence of a small negative correlation between age at emigration and the
fraction of physicians abroad. But it is not statistically signi…cant at the 5% level, and more importantly, it
suggests that a near tripling of the fraction of physicians abroad is correlated with a decline in the emigration
age of just 0.2 years. This suggests a miniscule bias to the earlier coe¢ cient estimates.
Rows 2 and 3 of the table use a survey of 1,752 physicians born in 31 African countries living in the
US and Canada in 2006, detailed in Clemens (2006). Row 2 uses survey questions regarding year of birth
and year of arrival in the US or Canada to create an unbiased estimate of the individual’s age at arrival
and repeats the analysis of row 1, …nding no evidence of a negative correlation between migration age and
diaspora size. Row 3 creates a dummy variable for each individual taking the value 1 if he or she arrived in
the US or Canada earlier than six years before earning his or her medical degree (about 18% of the sample).
Such a person probably did not go abroad for the express purpose of becoming a physician and may not
have decided upon a career at the moment of emigration. Here we would expect a strong positive coe¢ cient
if corr(li ; ci ) < 0, but there is no evidence of such a correlation.
Finally, row 4 uses the microdata 25% sample of the 1999 census of France, which reports survey forms
…lled out by 2,664 individual physicians born in 37 African countries. The year of arrival in France is not
recorded. But for countries of birth whose physicians typically emigrated to France later in life, after their
medical degrees (and thus highly likely to be physicians if they had not emigrated), the typical reported
age should be higher than that for countries of birth whose physicians typically emigrated prior to their
degree and thus enter the sample immediately upon graduation. We do see weak evidence of an extremely
small negative correlation between age in 1999 and the emigration indicators. But the coe¢ cients are not
statistically signi…cant, and they are tiny; a near tripling of emigration rates correlates with a decline in
34
average age of about 0.1 years.
7
Aside: The causes of health professional emigration
We note in passing that Tables 2 and 3 suggest the magnitudes of a few important determinants of health
professional emigration from Africa. Two of these are income per capita and the occurrence of a major war.
Doctors and nurses tend to leave richer African countries than poorer ones. The elasticity of physicians
per capita abroad with respect to sending-country income per capita is about 0.7 (Table 2), for nurses
about 0.5 (Table 3). Both are positive and statistically signi…cant at the 1% level. This is in line with
the evidence presented by Hatton and Williamson (2005) that in the poorest countries— and there alone—
economic development appears to encourage emigration by helping potential emigrants overcome …xed costs
associated with working overseas.
Lucas (2005, Ch. 3) argues that this evidence of an empirical “inverted-U” relationship between income
per capita and emigration is not robust across all countries and for the mass of migrants. But this is not
incompatible with the notion that of an inverted-U relationship for high-skill professions in the poorest
countries. In addition to overcoming …xed costs, potential high-skill migrants must acquire su¢ cient human
capital to compete in the global labor market. Economic development and its associated higher quality
training may assist in such competitiveness. Certainly the fact that South African nurses are in higher
demand internationally than their Sierra Leonean colleagues has to do with more than the ability of the
former to pay for airline travel; South African nursing schools have a superior ability to meet international
training standards. A case in point of this phenonmenon may be Rwanda, where strong economic recovery
and stability following the genocide have allowed the Kigali Health Institute to recently become the …rst
and only producer of fully-‡edged Registered Nurses in a country that traditionally has only trained preprofessional (“A2”) nurses. Given that Registered Nurses are in far higher demand abroad than vocational
nurses, it is quite possible that the emigration of Rwandan nurses could correspondingly increase.
The data make it clear that major wars are associated with a tremendous exodus of skilled health
professionals, a phenomenon that has not heretofore been systematically measured. “Major”war constitutes
more than 25 battle-related deaths per year and a total con‡ict history of more than 1000 battle-related
deaths. The sample average of the log of physicians per capita abroad is
0:569, which increases on average by
about 0:6 in countries that have experienced major warfare in the preceding thirty years (Table 2, regression
(9)). The change is statistically signi…cant at the 2% level. In other words, expected physician migration
35
nearly doubles, from about 0.6 physicians per 10,000 population to about 1.0, in countries touched by war.
The e¤ect is persistent across decades. For nurses, warfare is associated with a jump of nurses abroad per
10,000 population from 0.7 to 1.6, again roughly a doubled exodus. This di¤erence is statistically signi…cant
at the 1% level. To the other consequences of African wars we may add another: doubling the departure of
a country’s health care providers.
8
Conclusion
8.1
Empirical observations
Tables 2 and 3 suggest some lessons. First, we have a partial suggested answer to the question of whether
the positive correlation between health professional emigration and health professional stocks is driven by
positive causation from emigration to stocks, positive causation from stocks to emigration, or by third factors.
The …rst e¤ect— a positive e¤ect of migration on domestic stocks, for example as foreign opportunities
encourage young people to enter the health …eld— is not large enough to be the driving force. Instead,
the big determinant of the positive correlation is some combination of the e¤ects of large domestic stocks
in countries ill-equipped to absorb new physicians domestically (causation from stocks to emigration) and
national characteristics that raise production and emigration simultaneously (such as the quality of medical
schools).
Second, the negative e¤ect of emigration on domestic stocks, an accounting identity, is being counteracted
by some positive e¤ect of emigration on domestic stocks such that the net e¤ect is too small to be observed.
Since almost half the countries in question have lost more than 40% of their physicians, the e¤ect should
be easy to observe in the absence of a large countervailing force. To the extent that the instruments used
here are valid, and because the negative e¤ect of emigration on stocks is an identity, the existence of a
simultaneous strong positive e¤ect of migration on stocks is the only reason the e¤ect of migration on stocks
could be zero in Table 2. Analysis of this type does not illuminate the nature of this countervailing force. It
could be supply-driven, as African governments are typically able to increase the production of physicians
and nurses to replace those who have left. It could be demand-driven, as people enter the medical …eld in
greater numbers encouraged by the foreign success of their departed compatriots. Whatever it is, it has been
su¢ cient to totally counteract any sta¢ ng e¤ect from even the massive exoduses of health personnel seen in
Figures 2 and 3.
Third, we might have reasons to believe that emigration could degrade health outcomes or erode the
availability of care, even if the emigrants are replaced by new entrants to the profession. Emigration could
36
a¤ect the domestic distribution of health professionals, for example by pulling them into urban areas and
into the private sector to replace emigrants from those sectors. Emigration could also change the average
level of training or experience in the domestic workforce, as younger professionals with less training replace
those who are recruited abroad. This could have short term consequences for the quality of care and longterm consequences on professional pedagogy. But if these things are happening systematically across Africa,
we see no evidence that they substantially a¤ect the eleven indicators of mass primary care availibility and
public health outcomes tested here.
The empirical literature on the sending-country e¤ects of migration is in its infancy. We certainly need
better measures of migration— in particular, thoughtful agreement on the de…nition of a ‘migrant’— and we
need better natural (and even designed) experiments. This takes a stance on both issues, employing adequate
though problematic indicators of emigration and imperfect but provocative natural experiments. Much can
be learned, this paper claims, even with this ‡awed machinery. The …eld nevertheless remains wide open
to improved measures of worker movement and to creative experiments with greater internal and external
validity.
8.2
The missing link between migration and sending-country welfare: Health
systems
Many health economists …nd this quite intuitive, even if migration researchers do not. But if we seek to
set migration policy to maximize welfare, we must understand the health systems that translate human
resources into social welfare. The state of those systems is dire, but it is not at all clear that a dearth of
persons is a greater constraint than the incentives those persons face.
The domestic distribution of health professionals across the continent is already highly skewed away
from sectors relevant to the poor masses (Serneels et al. 2005). That is, only a small minority of the most
highly trained health professionals (the ones most competitive abroad) in essentially all African countries
spend a substantial portion of their time working in rural areas and slums. The number of trained health
professionals within South Africa and Kenya who work outside the public sector or entirely outside the
health sector greatly exceeds the number who work outside the country (e.g. Hall and Erasmus 2004, p.
537; Personal communication from the the National Nurses Association of Kenya, April 25. 2006). A study
of ten Tanzanian government health centers found that 26% of sta¤ time was wasted on “breaks, waiting for
patients, ... social contacts and ... unexplained absences” (Kurowski et al. 2004). In Uganda, unannounced
visits to 100 government-run health facilities found 37% of care providers were simply not there at a time
37
they were being paid to be working (Chaudhury et al. 2006). And that is in the clinics where people had
accepted jobs; many rural clinics across Africa can …nd no one at all to …ll them. But it is specious to
suggest, as many do, that such un…lled positions are primarily a function of the number of health workers
who are somewhere within the national borders. By and large, incentives are insu¢ cient to convince existing
workers to take those jobs, and it is unclear why this would be less true for a potential emigrant whose
pathway is blocked by visa caps or recruitment bans.
Moreover, the extensive skills of highly-trained primary care providers have may be very ill-suited to
address the principal causes of mass morbidity and mortality among the poor (e.g. Filmer, Hammer, and
Pritchett 2000, 2002; Canning 2006). Children do not die in rural Mozambique primarily due to a lack of
cardiologists and nurse practitioners; they die principally from lack of oral rehydration during diarrhea, lack
of malaria prophylaxis, and lack of basic primary treatment for acute respiratory infections. None of these
require highly trained personnel to deliver. This is true even at the heart of the HIV catastrophe; in Zambia
only 16% of child death is HIV related (WHO-AFRO 2006, p. 135). These e¤ects accord with the model
here; equation (9) suggests that the public welfare e¤ect of a health professional is smallest in countries where
domestic market failures are most pronounced, and that the impact of a highly-trained health professional is
smallest in countries where that person can be substituted by a less-trained professional capable of delivering
the same care or— better in many cases— working for prevention.
As Karl Popper and others have argued, it is impossible to prove a null hypothesis. The zero coe¢ cient
spanning the …rst row of many of these regression tables does not “prove” anything. But why is it that
no sta¢ ng or public health e¤ects due to emigration per se are observable even across countries that have
lost half or even two thirds of their health professionals to emigration? Perhaps it is that the world is
complex, and superimposing an additional market failure— an impermeable border— onto the myriad other
failures in developing countries does little to improve welfare in a second-, third-, or tenth-best economy.
Punishing emigration, restricting quotas, and banning recruitment, while as plausible as import-substituting
industrialization once was, and for similar reasons, may at best make no one better o¤ and at worst make
everyone worse o¤.
38
References
Aitken, Norman D. (1968), “The international ‡ow of human capital: Comment”, American Economic
Review 58 (3): 539-545.
Alesina, Alberto, Enrico Spolaore, and Romain Wacziarg (2005), “Trade, growth, and the size of countries”,
Chap. 23 in Philippe Aghion and Steven N. Durlauf, eds., Handbook of Economic Growth, I. B. (New York:
Elsevier), pp 1499-1542.
American Medical Association (2007), “Practicing Medicine
assn.org/ama/pub/category/10141.html], accessed January 26.
in
the
US” [http://www.ama-
Anderson, T.W. (1951), “Estimating linear restrictions on regression coe¢ cients for multivariate normal
distributions”, Annals of Mathematical Statistics, 22 (3): 327-351.
Anderson, T. W. and Herman Rubin (1949), “Estimation of the parameters of a single equation in a complete
system of stochastic equations”, Annals of Mathematical Statistics 20 (1): 46-63.
Awases, M., A. Gbary, J. Nyoni, and R. Chatora (2004), Migration of Health Professionals in Six Countries:
A Synthesis Report (Brazzaville, Rep. of Congo: World Health Organization Regional O¢ ce for Africa).
Barro, Robert J. and Jong-Wha Lee (2000), “International Data on Educational Attainment: Updates
and Implications”, CID Working Paper No. 42, Kennedy School of Government (Cambridge, MA: Harvard
University).
Beine, Michel, Frédéric Docquier, and Hillel Rapoport (2001), “Brain drain and economic growth: Theory
and evidence”, Journal of Development Economics 64 (1): 275-289.
Beine, Michel, Frédéric Docquier, and Hillel Rapoport (2003), “Brain drain and LDCs’ growth: Winners
and losers”, IZA Discussion Paper 819 (Bonn: Forschungsinstitut zur Zukunft der Arbeit).
Bhargava, Alok and Frédéric Docquier (2006), “HIV pandemic, medical brain drain and economic development in sub-Saharan Africa”, Working Paper, Dept. of Economics (Houston, Texas: University of Houston)
Canning, David (2006),“The Economics of HIV/AIDS in Low-Income Countries: The Case for Preventio”,
Journal of Economic Perspectives 20(3): 121-142.
Chaudhury, Nazmul, Je¤rey Hammer, Michael Kremer, Karthik Muralidharan, and F. Halsey Rogers (2006),
“Missing in Action: Teacher and healther worker absence in developing countries”, Journal of Economic
Perspectives 20(1): 91-116.
Clemens, Michael A. (2006), “The …nancial impact of African health worker emigration,”forthcoming Working Paper (Washington, DC: Center for Global Development).
Clemens, Michael A. and Gunilla Pettersson (2006), “A new database of health professional emigration from
Africa”, Working Paper 95 (Washington, DC: Center for Global Development).
Cragg, John G. and Stephen G. Donald (1993), “Testing identi…ability and speci…cation in instrumental
variables models”, Econometric Theory 9(2): 222-240.
Curtin, Philip, Steven Feierman, Leonard Thompson, and Jan Vansina (1978), African History (Boston:
Little, Brown and Co.).
Docquier, Frédéric and Hillel Rapoport (2004), “Skilled migration: The perspective of developing countries”,
World Bank Policy Research Working Paper 3382 (Washington, DC: World Bank).
Docquier, Frédéric and Abdeslam Marfouk (2005), “International Migration by Educational Attainment
(1990-2000) - Release 1.1”, update of World Bank Policy Research Working Paper 3381 (Washington, DC:
World Bank).
39
Dumont, Jean-Christophe, and Georges Lemaître (2005), “Counting immigrants and expatriates in OECD
countries: A new perspective,” Directorate for Employment Labour and Social A¤airs (Paris: OECD).
Educational Commission for Foreign Medical Graduates (2007), International Medical Education Directory
(IMED) [http://imed.ecfmg.org], accessed January 26.
Fage, J. D. (1978), A History of Africa (New York: Alfred A. Knopf).
Filmer, Deon, Je¤rey S. Hammer, and Lant H. Pritchett (2000), “Weak Links in the Chain: A diagnosis of
health policy in poor countries”, World Bank Research Observer 15(2): 199-224.
Filmer, Deon, Je¤rey S. Hammer, and Lant H. Pritchett (2002), “Weak Links in the Chain II: A prescription
for health policy in poor countries”, World Bank Research Observer 17(1): 47-66.
Frankel, Je¤rey A. and David Romer (1999), “Does trade cause growth?” American Economic Review 89
(3): 379-399.
Gallup, John Luke, Je¤rey D. Sachs, and Andrew Mellinger (1998), “Geography and Economic Development”
in Boris Pleskovic and Joseph E. Stiglitz, eds., Annual World Bank Conference on Development Economics
1998 (The World Bank: Washington, DC).
Gann, L. H. (1969), “Re‡ections on imperialism and the scramble for Africa”, in L. H. Gann and Peter
Duignan, eds., Colonialism in Africa, 1870-1960 (New York: Cambridge University Press).
Grubel, Herbert G. and Anthony D. Scott (1966), “The international ‡ow of human capital”, American
Economic Review 56 (1/2): 268-274.
Grubel, Herbert G. and Anthony D. Scott (1968), “The international ‡ow of human capital: Reply”, American Economic Review 58 (3): 545-548.
Hagopian, Amy, Matthew J. Thompson, Meredith Fordyce, Karin E. Johnson and L. Gary Hart, (2004),
“The migration of physicians from sub-Saharan Africa to the United States of America: Measures of the
African brain drain”, Human Resources for Health 2: 17.
Hall, Elsje and Johan Erasmus (2004), “Medical practitioners and nurses”, in HSRC, Human Resources
Development Review 2003 (Pretoria: Human Sciences Research Centre), pp.523-53.
Hamilton, Bob and John Whalley (1984), “E¢ ciency and distributional implications of global restrictions
on labor mobility”, Journal of Development Economics 14 (1): 61-75.
Hansen, Lars Peter. (1982), “Large Sample Properties of Generalized Method of Moments Estimators”,
Econometrica 50 (4): 1029-1054.
Harris, John R. and Michael P. Todaro (1970), “Migration, unemployment, and development: A two-sector
analysis”, American Economic Review 60 (1): 126-142.
Hatton, Timothy J. and Je¤rey G. Williamson (2003), “Demographic and Economic Pressure on Emigration
out of Africa”, Scandiavian Journal of Economics 105 (3): 465-486.
Hatton, Timothy J. and Je¤rey G. Williamson (2005), “What fundamentals drive world migration?”, in
George J. Borjas and Je¤ Crisp, eds., Poverty, International Migration and Asylum (New York: Palgrave
Macmillan), pp. 15-38.
Kurowski, Christoph, Kaspar Wyss, Salim Abdulla, N’Diekhor Yémadji and Anne Mills (2004), “Human
Resources for Health: Requirements and Availability in the Context of Scaling-Up Priority Interventions
in Low-Income Countries”, HEFP working paper 01/04 (London: London School of Hygiene and Tropical
Medicine).
La Porta, Rafael, Florencio López de Silanés, Andrei Shleifer, and Robert W. Vishny (1997), “Legal determinants of external …nance”, Journal of Finance 52 (3): 1131-1150.
40
Lucas, Robert E. B. (2005), International Migration Regimes and Economic Development (Northampton,
MA: Edward Elgar Publishing).
McEwan, P. J. M. (1968), Nineteenth-Century Africa (London: Oxford University Press).
Miyagiwa, Kaz (1991), “Scale economies in education and the brain drain problem”, International Economic
Review, 32 (3), 743-759.
Moses, Jonathon W. and Bjørn Letnes (2004), “The economic costs to international labor restrictions:
Revisiting the empirical discussion”, World Development 32 (10): 1609-1626.
Mountford, Andrew (1997), “Can a brain drain be good for growth in the source economy?” Journal of
Development Economics 53 (2): 287-303.
Mullan, Fitzhugh (2005), “The metrics of the physician brain drain”, New England Journal of Medicine
353:1810-1818.
Oliver, Roland and Anthony Atmore (2005), Africa Since 1800, 5th ed. (New York: Cambridge University
Press).
Oliver, Roland and J. D. Fage (1963), A Short History of Africa (New York: NYU Press).
Oosthuizen, Martha Johanna (2005), An analysis of the factors contributing to the emigration of South Africa
nurses, PhD dissertation, Department of Health Studies (Pretoria: University of South Africa).
Rodríguez, Francisco and Dani Rodrik (2001), “Trade policy and growth: A skeptic’s guide to the crossnational evidence”, in Benjamin Bernanke and Kenneth S. Rogo¤, eds. Macroeconomics Annual 2000 (Cambridge, MA: MIT Press).
Sanderson, G. N. (1985), “The European partition of Africa: Origins and Dynamics”, in Roland Oliver and
G. N. Sanderson, eds., The Cambridge History of Africa, Vol. 6 (New York: Cambridge University Press).
Serneels, Pieter, Magnus Lindelöw, José García Montalvo, Abigail Barr (2005), “For public service or money:
Understanding geographical imbalances in the health workforce”, Policy Research Working Paper 3686
(Washington, DC: World Bank).
Stark, Oded, Christian Helmenstein, and Alexia Prskawetz (1997), “A brain gain with a brain drain”,
Economics Letters 55 (2): 227-234.
Stark, Oded, Christian Helmenstein, and Alexia Prskawetz (1998), “Human capital depletion, human capital
formation, and migration: a blessing or a ‘curse’?” Economics Letters 60 (3): 363-367.
Stark, Oded and Yong Wang (2002), “Inducing human capital formation: migration as a substitute for
subsidies”, Journal of Public Economics 86 (1): 29-46.
Stock, James H., Jonathan H. Wright, and Motohiro Yogo (2002), “A survey of weak instruments and
weak identi…cation in generalized method of moments”, Journal of Business and Economic Statistics 20 (4):
518-529.
Stock, James H. and Motohiro Yogo (2005), “Testing for weak instruments in linear IV regression”, in J. H.
Stock and D. W. K. Andrews (eds.), Identi…cation and Inference for Econometric Models: Essays in Honor
of Thomas J. Rothenberg (New York: Cambridge University Press).
Touval, Saadia (1966), “Treaties, borders, and the partition of Africa”, Journal of African History 7(2):
279-293.
Walmsley, Terrie L. and L. Alan Winters (2005), “Relaxing the Restrictions on the Temporary Movement
of Natural Persons: A Simulation Analysis”, Journal of Economic Integration,.20 (4): 688-726.
WHO (2006), World Health Report 2006 (Geneva: World Health Organization).
WHO-AFRO (2006), African Regional Health Report 2006 (Brazzaville: World Health Organization African
Regional O¢ ce).
41
MPLs
MPLr
K
H
G
ws0
J
A
w*
w*
F
wr0
wˆ
wˆ
B
E
I
0
s
wˆ
Os
C
D
Lˆ
L*
L0
Figure 1: A simple welfare analysis of skilled migration
42
Or
Figure 2: Fraction of physicians abroad, 2000
Mozambique
Guinea-Bissau
Angola
Liberia
Equatorial Guinea
São Tome & P.
Malawi
Zambia
Ghana
Gambia
Congo, Rep.
Tanzania
Senegal
Zimbabwe
Cape Verde
Kenya
Mauritius
Cameroon
Namibia
Algeria
Rwanda
Uganda
Sierra Leone
Central Afr. Rep.
Togo
Madagascar
Burundi
Eritrea
Benin
Lesotho
Tunisia
Somalia
Comoros
Morocco
Ethiopia
Seychelles
Swaziland
Djibouti
Mali
Chad
South Africa
Burkina Faso
Gabon
Côte d'Ivoire
Nigeria
Sudan
Mauritania
Botswana
Guinea
Congo, Dem.
Niger
Libya
Egypt
0.0
0.2
0.4
0.6
0.8
Fraction of physicians abroad, 2000
43
1.0
Figure 3: Fraction of professional nurses abroad, 2000
Liberia
Burundi
Gambia
Mauritius
Sierra Leone
São Tome & P.
Cape Verde
Eritrea
Equatorial Guinea
Seychelles
Madagascar
Senegal
Central Afr. Rep.
Guinea-Bissau
Zimbabwe
Ghana
Comoros
Togo
Cameroon
Mozambique
Ethiopia
Malawi
Mali
Morocco
Rwanda
Benin
Angola
Congo, Dem.
Congo, Rep.
Nigeria
Chad
Uganda
Somalia
Zambia
Algeria
Kenya
Mauritania
Côte d'Ivoire
Guinea
Gabon
Namibia
Tunisia
South Africa
Tanzania
Swaziland
Lesotho
Niger
Burkina Faso
Botswana
Libya
Djibouti
Sudan
Egypt
0.0
0.2
0.4
0.6
0.8
Fraction of professional nurses abroad, 2000
44
1.0
Figure 4: Simple correlation between physicians per capita at home vs. abroad
15
EGY
10
ZAF
DZA MUS
TUN
5
MAR
1
SWZ
SDN DJI
MRT
NER
STP
NAM
COG
GNQ
GAB
BWA
NGA
GNB
KEN
CPV
GIN
GHA
CIV
MDG
AGO
SEN
COM
CMR
SLE
ZAR
ZMB
BEN
TGO
ZWE
BDI LSO
UGA
MLI
SOM
BFA
CAFGMB
ERI
ETH
TCD
TZA MOZ
LBR
RWA
MWI
.1
Physicians per 10,000 pop.
SYC
LBY
.1
1
5
Physicians abroad per 10,000 pop.
45
10 15
Figure 5: Simple correlation between child mortality rate and physicians abroad
300
SLE
NER
AGO
LBR
ERI
SDN
100
200
ZAR
ZMB
MOZ
BFATCD
BDI MWI
RWA
MRT CAF NGA
GIN
ETH
CIV
TZA
BEN
CMR
DJISWZ
TGO
MDG
SEN
LSO
GMB
UGA
BWA
GAB
COM
GNB
GNQ
KEN
ZWE
GHA
COG
STP
ZAF
NAM
EGY
MAR
LBY
DZA
TUN
CPV
MUS
SYC
0
Child mortality rate
MLI
SOM
.1
1
Physicians abroad per 10,000 pop.
46
10
Table 1: Summary statistics
ln(var)
Variable
N
Min
Max
Physicians abroad/10,000 pop.
(by country of birth)
Physicians abroad/10,000 pop.
(by country of training)*
Physician emigration/stock
Physicians/10,000 pop
Physician destination wage
Physician destination GDP/cap.
53
1.18
1.65
0.034
6.93
–0.569
1.23
53
0.249
0.598
0.000
3.72
–2.69
1.65
53
53
53
53
0.369
2.51
86,600
26,900
0.193
4.07
18,600
6,890
0.048
0.110
39,300
7,250
0.759
22.2
129,000
36,000
–1.17
0.088
11.3
10.2
0.649
1.24
0.234
0.354
Nurses abroad/10,000 pop
Nurse emigration/stock
Nurses/10,000 pop
Nurse destination wage
Nurse destination GDP/cap.
53
53
53
53
53
2.28
0.164
11.5
36,300
29,400
5.98
0.165
15.2
5,320
6,140
0.0528
0.00582
0.880
24,300
9,540
38.2
0.717
86.2
47,000
37,300
–0.302
–2.33
1.82
10.5
10.3
1.35
1.13
1.09
0.148
0.278
GDP per capita (2000 US$)
Primary school enrollment (%)
Land area, million km2
Infant mortality/1000
Child mortality/1000
Measles vacc. (fraction <1)
DPT vacc. (% age 1-2)
ARI prevalence (% <5)
ARI treatment (% <5)
Births attended (%)
Diarrhea prevalence (% <5)
Diarrhea treatment (% <5)
HIV prevalence (% 15-49)
% adults with HIV on ART
Francophone
Armed con‡ict 1970-99
53
53
53
53
53
53
53
29
29
46
31
30
42
35
53
53
973
89.4
0.554
90.5
142
0.655
64.4
19.3
40.8
53.6
19.0
29.5
8.26
9.19
0.396
0.434
1,390
27.7
0.627
39.4
69.1
0.198
22.1
12.6
17.4
22.3
7.47
12.9
9.69
19.9
0.494
0.500
89.0
17.0
0.000450
9.90
14.0
0.190
24.4
3.70
14.2
5.60
5.75
12.0
0.100
0
0
0
7,620
144
2.382
180
316
0.970
98.2
43.9
69.1
98.5
40.0
54.9
38.2
100
1
1
6.20
4.43
–1.74
4.37
4.78
–0.477
4.10
2.75
3.61
3.87
2.86
3.29
1.38
1.33
–
–
1.11
0.389
2.11
0.605
0.696
0.350
0.387
0.678
0.473
0.532
0.435
0.457
1.45
1.24
–
–
* The Bhargava and Docquier (2006) measure of physician emigration focuses on country of training for measuring
emigration to principal destination countries, but uses country of citizenship or birth for other destination countries.
47
Table 2: The e¤ect of physicians abroad on domestic physician stock
Dependent variable:
ln Physicians/cap.
ln Physicians abroad/cap.
(country of birth)
ln Physicians abroad/cap.
(country of training)
ln GDP per capita
(1)
OLS
0.69
(8.61)
Dependent variable:
ln Physicians abroad/cap.
Francophone
ln Land area
0.36
(3.16)
0.03
(0.12)
(6)
2SLS
0.08
(0.30)
0.48
(3.45)
0.94
(4.09)
–0.21
(0.71)
0.27
(1.11)
–4.92
(1.92)
0.91
(4.13)
–0.26
(0.88)
0.25
(1.03)
–4.47
(1.77)
53
0.465
<0.001
53
0.702
<0.001
53
0.646
<0.001
53
0.661
<0.001
(3)
OLS
(4)
OLS
(7)
1st stg.
(8)
1st stg.
–0.76
(2.36)
–0.19
(2.64)
–0.90
(2.35)
–0.17
(2.28)
–1.84
(0.94)
0.68
(0.49)
–0.45
(1.75)
–0.16
(2.56)
0.59
(4.98)
0.73
(2.76)
0.59
(2.33)
–7.80
(6.27)
–0.24
(0.78)
–0.16
(2.75)
1.39
(0.90)
–1.03
(0.94)
0.60
(5.15)
0.77
(2.88)
0.62
(2.46)
–13.45
(1.52)
53
0.628
0.221
<0.001
22.65
0.001
6.66
0.974
0.808
53
0.636
0.237
<0.001
17.00
0.006
3.49
0.615
0.306
Con‡ict
N
Adj. R 2
p(coe¤s. zero)
(5)
2SLS
0.75
(6.26)
–0.56
(2.07)
0.13
(0.61)
–1.91
(1.33)
ln Primary enrollment
Constant
(2)
OLS
ln Physician wage abroad
ln Destination GDP/cap.
ln GDP/cap.
ln Primary enrollment
Con‡ict
Constant
–0.60
(2.91)
13.47
(1.29)
N
Adj. R2
Shea R2 of excl. instruments
p(coe¤s. zero)
F statistic
Anderson CC test, p-value
Cragg-Donald Statistic
Anderson-Rubin test, p-value
Hansen J-test, p-value
53
0.201
53
0.201
0.006
5.64
0.027
3.02
(9)
3SLS
0.12
(0.43)
(10)
3SLS
(11)
3SLS
0.13
(0.55)
–2.33
(0.92)
0.92
(5.50)
–0.53
(1.50)
0.22
(1.01)
–3.33
(1.44)
0.09
(0.70)
0.91
(6.71)
–0.29
(1.06)
0.18
0.82
–4.08
(2.34)
53
0.666
<0.001
53
0.695
<0.001
51
0.705
<0.001
–0.76
(2.59)
–0.20
(2.87)
–0.51
(1.83)
–0.17
(3.20)
–0.05
(0.03)
–0.34
(0.36)
0.68
(6.44)
–1.22
(3.04)
–0.16
(2.18)
1.65
(0.81)
–0.91
(0.64)
0.78
(5.25)
–0.61
(2.70)
0.61
(2.54)
–1.14
(0.14)
0.82
(2.43)
–17.12
(1.48)
53
0.232
53
0.594
51
0.554
<0.001
<0.001
<0.001
0.73
(4.17)
–0.46
(1.18)
Abs. val. of robust t-statistics in parentheses. “Anderson CC” is Anderson’s canonical correlation test for
weak instruments. Columns 7 and 8 are the …rst-stage regressions for columns 5 and 6. For 3SLS regressions, p(all
coe¢ cients zero) is 2 test. For others, F-test.
48
Table 3: The e¤ect of professional nurses abroad on domestic professional nurse stock
Dependent variable:
ln Nurses/cap.
ln Nurses abroad/cap.
(country of birth)
ln GDP per capita
(1)
OLS
0.34
(3.56)
Dependent variable:
ln Nurses abroad/cap.
Francophone
ln Land area
(6)
2SLS
(9)
3SLS
1.93
(13.4)
0.07
(0.49)
0.79
(7.01)
–0.22
(0.84)
0.20
(1.09)
–2.15
(1.52)
–1.60
(1.00)
0.07
(0.52)
0.79
(7.37)
–0.25
(0.78)
0.20
(0.99)
–2.05
(1.31)
53
0.157
<0.001
53
0.595
<0.001
53
0.594
<0.001
53
0.594
<0.001
53
0.617
<0.001
53
0.625
<0.001
(3)
OLS
(4)
OLS
(7)
1st stg.
(8)
1st stg.
–0.66
(2.09)
–0.33
(4.11)
–1.04
(1.78)
–0.33
(4.10)
–1.91
(0.75)
0.71
(0.61)
–0.36
(1.28)
–0.33
(4.60)
–0.63
(2.09)
–0.33
(4.73)
0.37
(2.60)
0.78
(2.67)
–6.80
(4.74)
–0.60
(1.21)
–0.33
(4.61)
–1.16
(0.59)
0.60
(0.65)
0.36
(2.52)
0.78
(2.69)
–0.64
(0.04)
–0.64
(2.70)
–0.71
(1.47)
–0.36
(5.47)
–1.26
(0.63)
0.56
(0.61)
0.48
(3.95)
0.78
(2.80)
3.45
(0.25)
53
0.570
0.373
<0.001
13.81
<0.001
13.96
0.389
0.151
53
0.573
0.378
<0.001
10.18
<0.001
6.84
0.047
0.088
53
0.363
53
0.538
<0.001
<0.001
ln Nurse wage abroad
ln Destination GDP/cap.
ln GDP per capita
Con‡ict
Constant
–0.61
(2.71)
12.29
(0.71)
N
Adj. R2
Shea R2 of excl. instruments
p(coe¤s. zero)
F statistic
Anderson CC test, p-value
Cragg-Donald Statistic
Anderson-Rubin test, p-value
Hansen J-test, p-value
53
0.338
53
0.319
<0.001
11.16
0.001
5.70
0.07
(0.46)
0.75
(7.12)
–0.27
(0.83)
(10)
3SLS
0.07
(0.46)
0.79
(7.00)
–0.22
(0.83)
0.20
(1.07)
–2.15
(1.49)
Con‡ict
N
Adj. R 2
p(coe¤s. zero)
(5)
2SLS
0.09
(1.02)
0.78
(7.25)
–0.26
(0.90)
0.19
(0.98)
–1.91
(1.37)
ln Primary enrollment
Constant
(2)
OLS
Absolute value of robust t-statistics in parentheses. “Anderson CC” is Anderson’s canonical correlation test for
weak instruments. Columns 7 and 8 are the …rst-stage regressions for columns 5 and 6. For 3SLS regressions, p(all
coe¢ cients zero) is 2 test. For others, F-test.
49
Table 4: The e¤ect of physicians abroad on child mortality
Dependent variable:
ln Child mortality rate
(1)
OLS
(2)
OLS
(5)
2SLS
(6)
2SLS
(9)
3SLS
(10)
3SLS
ln Physicians abroad/cap.
(country of birth)
ln Physicians abroad/cap.
(country of training)
ln GDP per capita
–0.39
(6.19)
–0.21
(3.34)
–0.30
(2.20)
–0.26
(2.05)
–0.28
(1.98)
–0.25
(1.86)
–0.28
(2.79)
0.17
(0.91)
0.08
(0.55)
5.60
(4.06)
–0.31
(3.14)
0.12
(0.74)
0.07
(0.46)
5.96
(4.62)
–0.32
(3.16)
0.12
(0.53)
0.04
(0.33)
6.09
(4.17)
–0.31
(3.08)
0.08
(0.39)
0.06
(0.46)
6.13
(4.36)
–0.14
(2.03)
–0.36
(4.74)
–0.04
(0.21)
0.09
(0.74)
6.74
(6.65)
53
0.648
<0.001
53
0.655
<0.001
51
0.704
<0.001
–0.72
(2.38)
–0.20
(2.83)
–0.41
(1.36)
–0.19
(3.35)
0.67
(0.43)
–0.61
(0.58)
0.70
(6.61)
–1.06
(2.48)
0.16
(1.99)
2.71
(1.24)
–1.60
(1.04)
0.80
(5.35)
0.63
(2.69)
0.62
(2.58)
–6.79
(0.80)
0.82
(2.42)
–22.49
(1.80)
53
0.232
53
0.596
51
0.556
<0.001
<0.001
<0.001
Constant
4.56
(46.99)
–0.34
(3.81)
0.07
(0.65)
0.04
(0.33)
6.40
(7.55)
N
Adj. R 2
p(coe¤s. zero)
53
0.472
<0.001
53
0.629
<0.001
53
0.617
<0.001
53
0.625
<0.001
Dependent variable:
ln Physicians abroad/cap.
(3)
OLS
(4)
OLS
(7)
1st stg.
(8)
1st stg.
Francophone
–0.76
(2.36)
–0.19
(2.64)
–0.90
(2.35)
–0.17
(2.28)
–1.84
(0.94)
0.68
(0.49)
–0.49
(1.75)
–0.16
(2.56)
0.59
(4.98)
0.73
(2.76)
0.59
(2.33)
–7.80
(6.27)
–0.24
(0.78)
–0.16
(2.75)
1.39
(0.90)
–1.03
(0.94)
0.60
(5.15)
0.77
(2.88)
0.62
(2.46)
–13.45
(1.52)
53
0.628
0.221
<0.001
22.65
0.001
6.66
0.086
0.204
53
0.636
0.237
<0.001
17.00
0.006
3.49
0.053
0.060
ln Primary enrollment
Con‡ict
ln Land area
ln Physician wage abroad
ln Destination GDP/cap.
ln GDP/cap.
ln Primary enrollment
Con‡ict
Constant
–0.60
(2.91)
13.47
(1.29)
N
Adj. R 2
Shea R2 of excl. instruments
p(coe¤s. zero)
F statistic
Anderson CC test, p-value
Cragg-Donald Statistic
Anderson-Rubin test, p-value
Hansen J-test, p-value
53
0.201
53
0.201
0.006
5.64
0.027
3.02
(11)
3SLS
Absolute value of robust t-statistics in parentheses. “Anderson CC” is Anderson’s canonical correlation test for
weak instruments. Columns 7 and 8 are the …rst-stage regressions for columns 5 and 6. For 3SLS regressions, p(all
coe¢ cients zero) is 2 test. For others, F-test.
50
Table 5: The e¤ect of professional nurses abroad on child mortality
Dependent variable:
ln Child mortality rate
(1)
OLS
(2)
OLS
(5)
2SLS
(6)
2SLS
(9)
3SLS
(10)
3SLS
ln Nurses abroad/cap.
(country of birth)
ln GDP per capita
–0.27
(3.78)
4.70
(51.19)
–0.13
(2.22)
–0.41
(5.08)
0.01
(0.15)
0.00
(0.01)
7.24
(9.84)
–0.20
(2.62)
–0.39
(4.76)
0.11
(0.88)
0.03
(0.21)
6.65
(7.65)
–0.22
(2.92)
–0.38
(4.54)
0.13
(1.07)
0.04
0.26
6.50
(7.27)
–0.19
(2.24)
–0.42
(6.10)
0.05
(0.24)
–0.03
(0.21)
7.10
(7.11)
–0.20
(2.37)
0.37
(5.34)
0.02
(0.11)
0.03
(0.22)
6.91
(6.94)
N
Adj. R 2
p(coe¤s. zero)
53
0.269
53
0.603
53
0.589
53
0.582
53
0.619
53
0.617
Dependent variable:
ln Nurses abroad/cap.
(3)
OLS
(4)
OLS
(7)
1st stg.
(8)
1st stg.
Francophone
–0.66
(2.09)
–0.33
(4.11)
–1.04
(1.78)
–0.33
(4.10)
–1.91
(0.75)
0.71
(0.61)
–0.36
(1.28)
–0.33
(4.60)
–0.63
(2.10)
–0.33
(4.72)
0.37
(2.60)
0.78
(2.00)
0.78
(2.00)
–6.80
(4.74)
–0.60
(1.21)
–0.33
(4.61)
–1.16
(0.59)
0.60
(0.65)
0.36
(2.52)
0.78
(1.93)
0.78
(2.69)
–0.64
(0.04)
–0.84
(1.80)
–0.36
(5.53)
–1.99
(1.03)
1.00
(1.11)
0.48
(3.95)
–0.63
(2.71)
0.79
(2.82)
6.76
(0.50)
53
0.570
<0.001
14.28
<0.001
13.96
0.086
0.562
53
0.573
<0.001
10.18
<0.001
6.84
0.065
0.451
53
0.363
53
0.536
ln Primary enrollment
Con‡ict
Constant
ln Land area
ln Nurse wage abroad
ln Destination GDP/cap.
ln GDP/cap.
ln Primary enrollment
Con‡ict
Constant
–0.61
(2.71)
12.29
(0.71)
N
Adj. R 2
p(coe¤s. zero)
F statistic
Anderson CC test, p-value
Cragg-Donald Statistic
Anderson-Rubin test, p-value
Hansen J-test, p-value
53
0.338
53
0.319
Absolute value of robust t-statistics in parentheses. “Anderson CC” is Anderson’s canonical correlation test for
weak instruments. The 2SLS regression in column 5 uses the exogenous instruments in column 3, and the 2SLS
regression in column 6 uses the expanded instrument set in column 4. For 3SLS regressions, p(all coe¢ cients zero) is
2
test. For others, F-test.
51
Table 6: The e¤ect of physicians abroad on other public health indicators
Dependent variable:
ln Physicians abroad/cap.
(country of birth)
ln GDP per capita
ln Primary enrollment
Con‡ict
Constant
N
Adj. R 2
p(coe¤s. zero)
(1)
ln
IMR
(2)
ln
Measles
(3)
ln
DPT
(4)
ln ARI
prev.
(5)
ln ARI
treat.
–0.22
(1.74)
–0.25
(2.70)
0.06
(0.31)
0.06
(0.52)
5.48
(4.19)
0.12
(1.08)
–0.05
(0.61)
0.23
(1.30)
–0.02
(0.15)
–1.11
(0.98)
0.20
(1.78)
–0.08
(0.95)
0.29
(1.71)
–0.17
(1.60)
3.46
(3.09)
–0.26
(1.57)
0.07
(0.43)
1.15
(2.43)
0.20
(0.82)
–3.07
(1.37)
0.23
(2.15)
0.02
(0.16)
–0.33
(1.07)
–0.01
(0.06)
5.21
(3.54)
53
0.606
<0.001
53
0.118
0.018
53
0.279
<0.001
29
0.255
0.128
29
0.323
0.279
–0.43
(1.41)
–0.19
(3.40)
0.55
(0.36)
–0.50
(0.48)
0.70
(6.60)
0.62
(2.58)
–6.57
(0.78)
–0.47
(1.59)
–0.18
(3.17)
0.63
(0.42)
–0.60
(0.59)
0.70
(6.59)
0.60
(2.49)
–6.40
(0.77)
–0.51
(1.78)
–0.18
(3.34)
0.20
(0.14)
–0.30
(0.30)
0.69
(6.55)
0.60
(2.53)
–4.46
(0.55)
–0.96
(2.97)
–0.17
(2.04)
0.95
(0.53)
–2.33
(1.55)
0.41
(2.32)
0.22
(0.75)
9.82
(0.88)
–0.91
(2.88)
–0.16
(2.04)
0.79
(0.46)
–2.44
(1.64)
0.39
(2.25)
0.21
(0.73)
12.71
(1.16)
53
0.596
<0.001
53
0.596
<0.001
53
0.596
<0.001
29
0.616
<0.001
29
0.614
<0.001
Dependent variable:
ln Physicians abroad/cap.
Francophone
ln Land area
ln Physician wage abroad
ln Destination GDP/cap.
ln GDP/cap.
Con‡ict
Constant
N
Adj. R 2
p(coe¤s. zero)
Each column shows a single 3SLS regression. Absolute value of robust t-statistics in parentheses. ‘IMR’is infant
mortality rate per thousand live births. ‘Measles’is the fraction of children under age 1 vaccinated against measles.
‘DPT’ is the percentage of children 12-23 months vaccinated against DPT. ‘ARI prev.’ is the prevalence of acute
respiratory infections in children under 5 over a two week period requiring medical care, and ‘ARI treat’is the % of
those who saw a professional health care provider.
52
Table 7: The e¤ect of physicians abroad on other public health indicators (continued)
Dependent variable:
ln Physicians abroad/cap.
(country of birth)
ln GDP per capita
ln Primary enrollment
Con‡ict
Constant
N
Adj. R 2
p(coe¤s. zero)
(1)
ln Births
attend.
(2)
ln Diarr.
prev.
(3)
ln Diarr.
treat.
(4)
ln HIV
prev.
(5)
ln HIV
treat.
0.10
(0.90)
0.23
(2.62)
0.14
(0.75)
–0.17
(1.40)
1.99
(1.67)
–0.16
(1.02)
–0.17
(1.80)
–0.03
(0.18)
0.22
(1.54)
3.84
(2.91)
0.79
(2.49)
–0.35
(1.84)
–0.10
(0.30)
–0.29
(0.99)
6.64
(2.59)
1.41
(0.63)
–1.24
(0.87)
0.79
(0.33)
–0.75
(0.53)
6.89
(0.33)
0.54
(0.78)
0.08
(0.15)
0.37
(0.43)
–0.57
(1.13)
–0.17
(0.03)
46
0.597
<0.001
31
0.486
<0.001
30
–0.937
0.005
42
–0.805
0.051
33
0.239
0.003
–0.69
(2.11)
–0.18
(2.84)
0.20
(0.13)
–0.11
(0.10)
0.72
(5.79)
0.49
(1.87)
–6.53
(0.75)
–0.71
(1.89)
–0.11
(1.17)
–0.05
(0.02)
–0.77
(0.33)
0.66
(4.20)
0.39
(1.25)
3.70
(0.30)
–0.69
(2.55)
–0.03
(0.54)
0.65
(0.40)
–1.66
(0.99)
0.64
(4.13)
0.32
(1.02)
5.16
(0.62)
–0.72
(3.03)
–0.01
(0.18)
–1.75
(1.87)
0.70
(1.10)
0.68
(5.60)
0.53
(2.24)
7.85
(1.54)
–0.98
(3.11)
–0.13
(2.33)
–1.89
(1.42)
1.42
(1.63)
0.60
(4.78)
0.43
(1.76)
2.54
(0.34)
45
0.589
<0.001
31
0.475
<0.001
30
0.436
<0.001
42
0.503
<0.001
32
0.657
<0.001
Dependent variable:
ln Physicians abroad/cap.
Francophone
ln Land area
ln Physician wage abroad
ln Destination GDP/cap.
ln GDP/cap.
Con‡ict
Constant
N
Adj. R 2
p(coe¤s. zero)
Each column shows a single 3SLS regression. Absolute value of robust t-statistics in parentheses. ‘Births attend.’
is the percentage of deliveries attended by personnel trained to give the necessary supervision, care, and advice to
women during pregnancy, labor, and the postpartum period; to conduct deliveries on their own; and to care for
newborns. ‘Diarr. prev.’ is the percentage of children under age …ve with diarrhea requiring medical attention
over a two-week period, and ‘diarr. treat’is the percentage of those who received either oral rehydration therapy or
increased ‡uids, with continued feeding. ‘HIV prev.’ is the percentage of people ages 15-49 who are infected with
HIV, and ‘HIV treat’is the percentage of adults with advanced HIV infection receiving antiretroviral treatment.
53
Table 8: The relationship between emigrant stock and age at migration
Dependent variable:
ln Physicians abroad/cap.
Unweighted
Weighted
ln Physician emigr./stock
Unweighted
Weighted
United States, census 2000: 614 physicians from 18 African countries
Age at immigration
0:07
0:10
0:04
(1:65)
(0:94)
(1:49)
N
18
18
18
0:22
(1:84)
18
United States and Canada, survey 2006: 1,752 physicians from 31 African countries
Age at immigration
0:01
0:08
0:00
0:03
(0:50)
(1:78)
(0:30)
(0:43)
N
31
31
31
31
United States and Canada, survey 2006: 1,752 physicians from 31 African countries
Migrated >6 yrs. before degree
0:63
3:37
0:29
0:75
(1:27)
(2:95)
(0:80)
(0:73)
N
31
31
31
31
France, census 1999: 2,664 physicians from 37 African countries
Age in 1999
0:01
0:11
(0:25)
(1:24)
N
37
37
0:01
(0:46)
37
0:04
(0:81)
37
OLS regressions with absolute values of robust t-statistics in parentheses. All regressions include constant
term, not reported. Columns 2 and 4 conduct the regressions with analytic weights in which the variance of
each observation is inversely proportional to the number of census or survey observations used to generate
the country-average used as the right-hand side variable.
54
A
A.1
Appendix: Data sources
African physicians and professional nurses, abroad and domestic
See Clemens and Pettersson (2006) for sources of the measures of health professionals abroad based on
country of birth. The alternative ‘mixed’measure of African physicians abroad, which focuses on country of
training in measuring emigration to key receiving countries, comes from Bhargava and Docquier (2006).
A.2
Domestic health worker stocks in receiving countries
Physician and professional nurse stocks per 1,000 population in the United Kingdom, United States, Australia, Canada, and France in the year 2000 come from OECD Health Statistics 2006. These are multiplied
by the population of each of those countries in 2000 (from the World Bank’s World Development Indicators 2006 ) to get the total health professional stock in 2000. To calculate the number of physicians in
Francophone Canada, the total number of Canadian physicians was multiplied by 27.7%— the fraction of
Canadian physicians who lived in Quebec in 1996, according to J. Roger Pitblado and Raymond W. Pong
(1999), “Geographic Distribution of Physicians in Canada”, Centre for Rural and Northern Health Research
(Sudbury, Ontario: Laurentian University), page 3-8, Table 3.1. To calculate the number of professional
nurses in Francophone Canada, the total number in Canada was multiplied by 25.2%— which is the fraction
of Canada’s nurses living in Quebec in 2001 according to Canadian Institute for Health Information (2002),
Supply and Distribution of Registered Nurses in Canada 2001 (Ottawa: CIHI), page 11. The total population of Francophone Canada is assumed to be that of Quebec (24.1% of the Canadian population, according
to the 2001 census of Canada) and Anglophone Canada is assumed to be the rest of the country.
A.3
Health care and health outcomes
Fraction of children 0-1 year vaccinated against measles 2001; infant mortality rate per 1,000 live births in
2000; child (under age 5) mortality in 2000 are from the World Bank Africa Database 2004 (Washington,
DC: World Bank). Acute Respiratory Infection (ARI) prevalence (children under 5); ARI treatment (% of
children under 5 taken to a health provider); Births attended by skilled health sta¤ (% of total); Diarrhea
prevalence (% of children under 5); Diarrhea treatment (% of children under 5 who received ORS packet);
Immunization, DPT (% of children ages 12-23 months); and Prevalence of HIV, total (% of population
aged 15-49) in 2000 are taken from World Development Indicators 2005 (Washington, DC: World Bank).
Percentage of adults with advanced HIV infection receiving antiretroviral treatment are from WHO (2004),
Investing in a Comprehensive Health Sector Response to HIV/AIDS (Geneva: World Health Organisation),
Annex 1, pp 61-62 for Angola, Burundi, Burkina Faso, Botswana, Central African Republic, Côte d’Ivoire,
Cameroon, Dem. Rep. Congo, Djibouti, Ethiopia, Ghana, Guinea, Kenya, Lesotho, Mozambique, Malawi,
Namibia, Nigeria, Rwanda, Sudan, Somalia, Swaziland, Tanzania, Uganda, South Africa, Zambia, and
Zimbabwe. They are taken from UNAIDS (2003), Progress Report on the Global Response to the HIV/AIDS
Epidemic (Geneva: Joint United Nations Programme on HIV/AIDS), Annex 7, p. 90 for Benin, Eritrea,
Gambia, Equatorial Guinea, Mauritius, Senegal, Sierra Leone, and Seychelles.
A.4
Other development indicators
GDP per capita (constant 2000 US$) and primary school enrollment (% gross) in 2000 are taken from World
Development Indicators 2005 (Washington, DC: World Bank). Fraction of higher schooling complete in
pop 15+, 1999 is taken from Robert Barro and Jong-Wha Lee (2000), “International Data on Educational
Attainment Updates and Implications,”Working Paper 7911 (Cambridge, MA: National Bureau of Economic
Research). Land area in millions of square kilometers in 2001 comes from the World Bank Africa Database
2004 (Washington, DC: World Bank).
A.5
Francophone countries
The Francophone countries are de…ned as Algeria, Burundi, Benin, Burkina Faso, Central African Republic,
Chad, Côte d’Ivoire, Congo (Rep.), Congo (Dem. Rep.), Comoros, Djibouti, Gabon, Guinea, Morocco,
55
Madagascar, Mali, Mauritania, Niger, Rwanda, Senegal, and Tunisia. Note that Cameroon is not counted
among these, due to the o¢ cial policy of English/French bilingualism in place since 1961 (especially important
in tertiary education).
A.6
Con‡ict
Armed con‡ict data come from the PRIO Armed Con‡ict Database, version 2.1, produced by the Centre
for the Study of Civil War at the International Peace Research Institute, Oslo (PRIO) and the Department
of Peace and Con‡ict Research, Uppsala University. The database is described in Nils Petter Gleditsch
et al. (2002), “Armed Con‡ict 1946–2001: A New Dataset”, Journal of Peace Research 39(5): 615–637,
and in Håvard Strand et al. (2005), Armed Con‡ict Dataset Codebook, Version 3-2005 (Oslo: International
Peace Research Institute [PRIO]). The con‡ict indicator takes the value “1” if there was an armed con‡ict
of intensity 2 or 3 (on the PRIO scale of 1 to 3) in that country between 1970 and 1999, and “0”otherwise.
An intensity of 2 or greater signi…es more than 25 battle-related deaths per year and a total con‡ict history
of more than 1000 battle-related deaths.
A.7
Destination country economic conditions
For each of the nine receiving countries, average annual general practitioner physician wage in PPP-adjusted
US dollars was obtained from the ERI Salary Assessor international salary database, maintained by Baker,
Thomsen Associates, a compensation and bene…ts consulting …rm. This was used to create a weighted
average destination-country physician salary for each sending country, weighted by the fraction of physicians
born in that sending country and living abroad who live in that receiving country. The result is an index
of the physician wage faced by the ‘typical’ physician emigrant from each sending country. An identical
procedure created the professional nurse destination country wage index using Registered Nurse wages from
Baker, Thomsen Associates. The destination-country GDP per capita index was constructed in a similar
fashion, simply using exchange-rate GDP per capita in 2000 US$ instead of wages— where the weights in the
physician regressions are relative physician stocks abroad, and the weights are relative nurse stocks in the
nurse regressions.
A.8
Survey of African-born Physicians in the United States and Canada
See Clemens (2006).
A.9
Census estimates of age and age at immigration of African-born physicians
in the United States and France
The census data on age and age at immigration for the analysis of African-born physicians in Table 8 were
taken from anonymized microdata samples for the United States (5% sample) and France (25% sample):
United States: U.S. Dept. of Commerce, Bureau of the Census. Census of Population and Housing
2000 [United States]: Public Use Microdata Sample: 5-percent sample [Computer …le]. ICPSR release.
Washington, DC: U.S. Dept. of Commerce, Bureau of the Census [producer], 2003. Ann Arbor, MI: Interuniversity Consortium for Political and Social Research [distributor], 2003.
France: LASMAS Institut du Longitudinal and Centre National de la Recherche Scienti…que, custom
extract from Recensement de la Population : …chier détail au 1/4 - France métropolitaine - 1999, prepared
January 2006.
*
*
56
*